Skip to content
Python

feat(gui): hover help on the Steps panel, and fixes from manual testing

Devika Vermaf5cb467cad35
diff --git a/surfaces/gui/src/api.ts b/surfaces/gui/src/api.ts
index c58fa5af..5e24a0fa 100644
--- a/surfaces/gui/src/api.ts
+++ b/surfaces/gui/src/api.ts
@@ -2382,6 +2382,8 @@ export interface ObservabilityStatus {
failed: number;
dropped: number;
last_error: string;
+ // Model calls outside any session today (through this machine's model address; F-4).
+ outside_today?: { calls: number; tokens: number };
}
export async function getObservabilitySettings(machineId?: string | null): Promise<ObservabilitySettings> {
@@ -4534,8 +4536,8 @@ export interface TurnIssues {
info: number;
seen?: boolean;
issues: TurnIssue[];
- fine: { check: string; text: string }[];
- not_checked: { check: string; text: string }[];
+ fine: { check: string; text: string; facts?: Record<string, any> }[];
+ not_checked: { check: string; text: string; facts?: Record<string, any> }[];
}
export interface RecordedStep {
diff --git a/surfaces/gui/src/components/Help.tsx b/surfaces/gui/src/components/Help.tsx
new file mode 100644
index 00000000..f5ccb969
--- /dev/null
+++ b/surfaces/gui/src/components/Help.tsx
@@ -0,0 +1,44 @@
+// Hover and keyboard-focus help for a label (observability F-6). A dotted underline says help
+// is there; the text opens below the label (above it near the bottom of the window), drawn in
+// a layer over the page so a scrolling panel never clips it.
+import { useId, useRef, useState, type ReactNode } from "react";
+import { createPortal } from "react-dom";
+
+const WIDTH = 280;
+
+export function Help({ text, children }: { text?: string; children: ReactNode }) {
+ const id = useId();
+ const ref = useRef<HTMLSpanElement>(null);
+ const [at, setAt] = useState<{ left: number; top: number; up: boolean } | null>(null);
+ if (!text) return <>{children}</>;
+ const show = () => {
+ const r = ref.current?.getBoundingClientRect();
+ if (!r) return;
+ const left = Math.max(8, Math.min(r.left, window.innerWidth - WIDTH - 8));
+ const up = r.bottom + 140 > window.innerHeight;
+ setAt({ left, top: up ? r.top - 6 : r.bottom + 6, up });
+ };
+ const hide = () => setAt(null);
+ return (
+ <span
+ ref={ref}
+ className="help"
+ tabIndex={0}
+ aria-describedby={at ? id : undefined}
+ onMouseEnter={show}
+ onMouseLeave={hide}
+ onFocus={show}
+ onBlur={hide}
+ onKeyDown={(e) => e.key === "Escape" && hide()}
+ data-testid="help"
+ >
+ {children}
+ {at && createPortal(
+ <span role="tooltip" id={id} className={"help-tip" + (at.up ? " up" : "")} style={{ left: at.left, top: at.top }} data-testid="help-tip">
+ {text}
+ </span>,
+ document.body,
+ )}
+ </span>
+ );
+}
diff --git a/surfaces/gui/src/components/IssueSentences.test.tsx b/surfaces/gui/src/components/IssueSentences.test.tsx
new file mode 100644
index 00000000..3eb89aa6
--- /dev/null
+++ b/surfaces/gui/src/components/IssueSentences.test.tsx
@@ -0,0 +1,50 @@
+// Issue sentences built on screen from their facts (TD2-3, fix list C-2) say what the engine
+// says in English, and leave no blank in Chinese. The fixture is every issue and fine line of
+// the 39 golden Phase 2 scenarios, as the engine's checks produce them.
+import { describe, expect, it } from "vitest";
+import i18n from "i18next";
+import en from "../locales/en.json";
+import zh from "../locales/zh.json";
+import golden from "./__fixtures__/issues-golden.json";
+import { fineSentence, issueSentence } from "./StepsPanel";
+
+const tFor = async (lng: "en" | "zh") => {
+ const inst = i18n.createInstance();
+ await inst.init({ lng, resources: { en: { translation: en }, zh: { translation: zh } }, interpolation: { escapeValue: false } });
+ return inst.t.bind(inst) as any;
+};
+
+// Where the screen's English deliberately reads better than the stored sentence.
+const READS_BETTER: Record<string, (stored: string, g: any, t: any) => string> = {
+ waited_unwatched: (s) => "It " + s,
+ shadow_disagreed: (s) => s.replace(/said (\w+) on/, (_m, r) => `said "${r === "allow" ? "allowed" : r === "deny" ? "blocked" : "asked you"}" on`),
+ // The error type in plain words ("key or sign-in refused", not "auth").
+ model_failed: (s, g, t) => s.replace(`(${g.facts.error_type})`, `(${t(`steps.error_type.${g.facts.error_type}`, { defaultValue: g.facts.error_type })})`),
+};
+
+describe("issue sentences from facts", () => {
+ it("say in English what the engine says, for every golden issue and fine line", async () => {
+ const t = await tFor("en");
+ expect(golden.length).toBeGreaterThan(300);
+ for (const g of golden as any[]) {
+ const said = g.kind === "issue" ? issueSentence(g, t) : fineSentence(g, t);
+ const want = g.kind === "issue" && READS_BETTER[g.check] ? READS_BETTER[g.check](g.text, g, t) : g.text;
+ expect(said, `${g.scenario} ${g.kind} ${g.check}`).toBe(want);
+ }
+ });
+
+ it("leave no blank in Chinese and never fall back to English", async () => {
+ const t = await tFor("zh");
+ for (const g of golden as any[]) {
+ const said = g.kind === "issue" ? issueSentence(g, t) : fineSentence(g, t);
+ expect(said, `${g.scenario} ${g.check}`).not.toBe(g.text);
+ expect(said).not.toMatch(/\{\{|undefined|null/);
+ }
+ });
+
+ it("keeps the stored English sentence for turns checked before facts were stored", async () => {
+ const t = await tFor("zh");
+ expect(issueSentence({ check: "tool_failed", text: "run_shell failed.", facts: { how: "exit" } }, t)).toBe("run_shell failed.");
+ expect(fineSentence({ check: "tool_failed", text: "No tool failed" }, t)).toBe("No tool failed");
+ });
+});
diff --git a/surfaces/gui/src/components/ObservabilitySection.tsx b/surfaces/gui/src/components/ObservabilitySection.tsx
index 2d0c03c7..da05c05a 100644
--- a/surfaces/gui/src/components/ObservabilitySection.tsx
+++ b/surfaces/gui/src/components/ObservabilitySection.tsx
@@ -240,6 +240,11 @@ export function ObservabilitySection({ machine }: { machine?: Machine | null })
? t("obs.status.failed_sub")
: t("obs.status.resumes"))}
</div>
+ {status?.outside_today?.calls ? (
+ <div className="text-meta text-muted mt-1" data-testid="obs-outside">
+ {t("obs.status.outside", { count: status.outside_today.calls, tokens: fmt(status.outside_today.tokens) })}
+ </div>
+ ) : null}
{status && (status.state === "retrying" || status.state === "failed") && status.last_error ? (
<div className="text-meta text-danger mt-1" data-testid="obs-status-error">
{status.last_error === "The destination refused the key" ? t("obs.refused_key") : status.last_error === "Address not found" ? t("obs.not_found") : status.last_error}
diff --git a/surfaces/gui/src/components/StepsPanel.test.tsx b/surfaces/gui/src/components/StepsPanel.test.tsx
index 0cca0bfa..44b6040b 100644
--- a/surfaces/gui/src/components/StepsPanel.test.tsx
+++ b/surfaces/gui/src/components/StepsPanel.test.tsx
@@ -2,7 +2,7 @@
import { cleanup, fireEvent, render, screen, waitFor } from "@testing-library/react";
import { afterEach, describe, expect, it, vi } from "vitest";
import { getRecordedTurns, getRecordedUsage, getStepDetail, getTurnIssues, getTurnSteps, markIssuesSeen } from "../api";
-import { StepsPanel, formatDuration } from "./StepsPanel";
+import { StepsPanel, formatDuration, stampLine } from "./StepsPanel";
import i18n from "i18next";
import zh from "../locales/zh.json";
@@ -362,6 +362,79 @@ describe("Gaps (OB-14)", () => {
});
});
+describe("Hover help (F-6)", () => {
+ it("explains a tile, a details label and the token legend on hover, and on keyboard focus", async () => {
+ vi.mocked(getRecordedTurns).mockResolvedValue([turn] as any);
+ vi.mocked(getTurnSteps).mockResolvedValue(steps as any);
+ vi.mocked(getStepDetail).mockResolvedValue({ ...steps[1], attrs: { ...steps[1].attrs, model: "openai:model-a", finish_reason: "tool_calls" } });
+ render(<StepsPanel sessionId="s1" onClose={() => {}} />);
+ await waitFor(() => expect(screen.getAllByTestId("steps-row")).toHaveLength(6));
+ // "Other" says what it is made of.
+ const other = screen.getAllByTestId("help").find((h) => h.textContent === "Other")!;
+ fireEvent.mouseEnter(other);
+ expect(screen.getByRole("tooltip").textContent).toContain("the session-title call");
+ fireEvent.mouseLeave(other);
+ expect(screen.queryByRole("tooltip")).toBeNull();
+ // Keyboard: focus shows it, Escape hides it.
+ const total = screen.getAllByTestId("help").find((h) => h.textContent === "Total")!;
+ fireEvent.focus(total);
+ expect(screen.getByRole("tooltip").textContent).toContain("How long the turn took");
+ expect(total.getAttribute("aria-describedby")).toBe(screen.getByRole("tooltip").id);
+ fireEvent.keyDown(total, { key: "Escape" });
+ expect(screen.queryByRole("tooltip")).toBeNull();
+ // A row's hover text names its kind.
+ expect(screen.getAllByTestId("steps-row")[1].getAttribute("title")).toContain("main work");
+ // Details: a label and a token part.
+ fireEvent.click(screen.getAllByTestId("steps-row")[1]);
+ await screen.findByTestId("steps-tokens");
+ const attempts = screen.getAllByTestId("help").find((h) => h.textContent === "Attempts")!;
+ fireEvent.mouseEnter(attempts);
+ expect(screen.getByRole("tooltip").textContent).toContain("sent to the provider");
+ fireEvent.mouseLeave(attempts);
+ const cache = screen.getAllByTestId("help").find((h) => (h.textContent || "").startsWith("from cache"))!;
+ fireEvent.mouseEnter(cache);
+ expect(screen.getByRole("tooltip").textContent).toContain("served from its cache");
+ });
+});
+
+describe("Fix list batch (4 Nov)", () => {
+ it("names a sandbox that enforces nothing instead of saying No sandbox (F-7)", () => {
+ const t = ((k: string, o?: any) => ({
+ "steps.sandbox.none": "No sandbox", "steps.sandbox.not_enforced": "not enforced", "steps.sandbox.full": "enforced",
+ } as Record<string, string>)[k] ?? o?.defaultValue ?? k) as any;
+ expect(stampLine({ sandbox: { provider: "runner-local", enforcement: "none" } }, t)).toBe("runner-local · not enforced");
+ expect(stampLine({ sandbox: { provider: "direct", enforcement: "none" } }, t)).toBe("No sandbox");
+ expect(stampLine({ sandbox: { provider: "openshell", enforcement: "full" } }, t)).toBe("openshell · enforced");
+ });
+
+ it("shows rules and AGENTS.md at 0, memories without ids, the audit log in order, and helper tiles", async () => {
+ const ctx = step({ step_id: "c", kind: "context", name: "context", duration_ms: 0,
+ attrs: { session_prompt_hash: "2bcb44", per_turn: { folders: 2, skills_menu: 0 }, size_chars: 654, block_hash: "1039e5",
+ session_start: { memories: [1, 2, 3], memory_mode: "index", rules: 0, agents_md: 0 } } });
+ const ex = step({ step_id: "e", kind: "tool", name: "explore", started_at: 1100, duration_ms: 3000 });
+ const hm = step({ step_id: "hm", kind: "model", name: "model", parent_step_id: "e", started_at: 1200,
+ attrs: { purpose: "helper", usage: { reported: true, input_fresh: 1000, cache_read: 500, cache_write: 0, output: 300 } } });
+ const tool = step({ step_id: "w", kind: "tool", name: "todo_write", call_id: "c9", started_at: 4200 });
+ vi.mocked(getRecordedTurns).mockResolvedValue([turn] as any);
+ vi.mocked(getTurnSteps).mockResolvedValue([ctx, ex, hm, tool] as any);
+ vi.mocked(getStepDetail).mockResolvedValueOnce(ctx as any).mockResolvedValueOnce({ ...tool, attrs: {}, audit: [
+ { stage: "finished" }, { stage: "started" }, { stage: "proposed" }] } as any);
+ render(<StepsPanel sessionId="s1" onClose={() => {}} />);
+ await waitFor(() => expect(screen.getAllByTestId("steps-row").length).toBeGreaterThan(1));
+ const tiles = screen.getByTestId("steps-tiles").textContent || "";
+ expect(tiles).toContain("1Helpers");
+ expect(tiles).toContain("1.8kHelper tokens");
+ fireEvent.click(screen.getAllByTestId("steps-row")[0]);
+ const details = await screen.findByTestId("steps-details");
+ await waitFor(() => expect(details.textContent).toContain("Your rules 0"));
+ expect(details.textContent).toContain("AGENTS.md files 0");
+ expect(details.textContent).toContain("Memories 3 · as an index");
+ expect(details.textContent).not.toContain("mem_");
+ fireEvent.click(screen.getAllByTestId("steps-row").find((r) => r.textContent?.startsWith("todo_write"))!);
+ expect((await screen.findByTestId("steps-audit")).textContent).toBe("3 entries for this call: proposed, started, finished");
+ });
+});
+
describe("formatDuration", () => {
it("reads like the mock", () => {
expect(formatDuration(100)).toBe("0.1s");
@@ -402,7 +475,11 @@ describe("Issues tab (OB-3)", () => {
expect(screen.queryByTestId("steps-fine")?.textContent ?? "").not.toContain("No step repeated");
fireEvent.click(screen.getByTestId("steps-fine-toggle"));
expect(screen.getByTestId("steps-fine").textContent).toContain("No step repeated 3 times in a row");
- expect(screen.getByTestId("steps-fine").textContent).toContain("No plan to check against");
+ expect(screen.getByTestId("steps-fine").textContent).not.toContain("plan");
+ // F-9: every check is accounted for; what did not apply is listed, with why.
+ expect(screen.getByTestId("steps-issues-summary").textContent).toBe("4 checks: 2 found something · 1 fine · 1 didn't apply");
+ fireEvent.click(screen.getByTestId("steps-na-toggle"));
+ expect(screen.getByTestId("steps-na").textContent).toBe("– Left its plan: no to-do list in this turn");
fireEvent.click(screen.getAllByText("Go to step")[0]);
await waitFor(() => expect(screen.getByTestId("steps-timeline")).toBeTruthy());
expect(screen.getAllByTestId("steps-row-issue").length).toBe(1);
diff --git a/surfaces/gui/src/components/StepsPanel.tsx b/surfaces/gui/src/components/StepsPanel.tsx
index ec30215b..94a8797d 100644
--- a/surfaces/gui/src/components/StepsPanel.tsx
+++ b/surfaces/gui/src/components/StepsPanel.tsx
@@ -7,6 +7,7 @@
import { Fragment, useEffect, useMemo, useState } from "react";
import { useTranslation } from "react-i18next";
+import { Help } from "./Help";
import {
getRecordedTurns,
getRecordedUsage,
@@ -85,9 +86,12 @@ export function stampLine(stamp: Record<string, any> | undefined, t: T): string
}
const sb = stamp.sandbox;
if (sb) {
- parts.push(!sb.enforcement || sb.enforcement === "none" || sb.provider === "direct"
+ // F-7: "No sandbox" only when none ran. A provider that confines nothing (runner-local)
+ // is named, with "not enforced", so the header agrees with the session's sandbox pill.
+ parts.push(!sb.provider || sb.provider === "direct"
? t("steps.sandbox.none")
- : [sb.provider, t(`steps.sandbox.${sb.enforcement}`, { defaultValue: sb.enforcement }),
+ : [sb.provider,
+ !sb.enforcement || sb.enforcement === "none" ? t("steps.sandbox.not_enforced") : t(`steps.sandbox.${sb.enforcement}`, { defaultValue: sb.enforcement }),
sb.profile ? t("steps.sandbox.network", { profile: sb.profile }) : ""].filter(Boolean).join(" · "));
}
if (stamp.persona_fp) parts.push(`ⓕ${stamp.persona_fp}`);
@@ -255,6 +259,22 @@ const KIND_CLASS: Record<string, string> = {
sandbox: "sbx",
};
+/** A hover help text (observability F-6), from `steps.help`. */
+function H(t: T, key: string): string {
+ return t(`steps.help.${key}`, { defaultValue: "" });
+}
+
+/** The hover help for a step's kind: the row's hover text and its details heading. */
+function kindHelp(step: { kind: string; name?: string; shadow?: boolean; attrs?: any }, t: T): string {
+ const a = step.attrs || {};
+ const key = step.kind === "model" ? `kind.model_${a.purpose || "main"}`
+ : step.kind === "auto_approver" ? (step.shadow ? "kind.shadow" : "kind.auto_approver")
+ : step.kind === "summary" ? (step.name === "trim" ? "kind.trim" : "kind.summary")
+ : step.kind === "notice" ? `notice.${step.name}`
+ : `kind.${step.kind}`;
+ return H(t, key);
+}
+
const GAPS = ["controller_offline", "nobody_watching", "clock_jump"];
/** A gap notice with a duration (OB-14): drawn as a dashed band, not a point. */
@@ -275,6 +295,21 @@ function clockTime(ms: number): string {
return new Date(ms).toLocaleTimeString([], { hour: "2-digit", minute: "2-digit" });
}
+/** Helpers this turn ran: explore calls with a helper's own model calls under them (F-8). */
+function helperCount(steps: RecordedStep[]): number {
+ const parents = new Set(steps.filter((s) => s.kind === "model" && s.attrs?.purpose === "helper" && s.parent_step_id).map((s) => s.parent_step_id));
+ return parents.size;
+}
+
+/** The helpers' tokens (input of every kind plus output), already inside the session's total. */
+function helperTokens(steps: RecordedStep[]): number {
+ return steps.filter((s) => s.kind === "model" && s.attrs?.purpose === "helper" && s.attrs?.usage?.reported)
+ .reduce((n, s) => {
+ const u = s.attrs.usage;
+ return n + (u.input_fresh || 0) + (u.cache_read || 0) + (u.cache_write || 0) + (u.output || 0);
+ }, 0);
+}
+
/** How long the turn spent with the machine offline, or with nobody watching (mock ob14). */
function gapTotal(steps: RecordedStep[], name: string): number {
return steps.filter((s) => s.kind === "notice" && s.name === name).reduce((n, s) => n + (s.duration_ms || 0), 0);
@@ -413,17 +448,17 @@ export function StepsPanel({
const picker = turns && turns.length > 0 && turnIdx != null && (
<span className="steps-turn-picker">
<button aria-label={t("steps.prev_turn")} disabled={turnIdx <= 0} onClick={() => setTurnIdx(turnIdx - 1)}>‹</button>
- <span data-testid="steps-turn">{t("steps.turn_of", { n: turnIdx + 1, total: turns.length })}</span>
+ <span data-testid="steps-turn"><Help text={H(t, "turn")}>{t("steps.turn_of", { n: turnIdx + 1, total: turns.length })}</Help></span>
<button aria-label={t("steps.next_turn")} disabled={turnIdx >= turns.length - 1} onClick={() => setTurnIdx(turnIdx + 1)}>›</button>
</span>
);
const timeline = turn && (
<div className="steps-timeline" data-testid="steps-timeline">
- <div className="steps-axis" aria-hidden>
- <span>{t("steps.axis.step")}</span>
- <span className="steps-ticks">{ticks(total).map((x, i) => <span key={i}>{x}</span>)}</span>
- <span className="steps-axis-time">{t("steps.axis.time")}</span>
+ <div className="steps-axis">
+ <span><Help text={H(t, "axis.step")}>{t("steps.axis.step")}</Help></span>
+ <span className="steps-ticks" aria-hidden>{ticks(total).map((x, i) => <span key={i}>{x}</span>)}</span>
+ <span className="steps-axis-time"><Help text={H(t, "axis.time")}>{t("steps.axis.time")}</Help></span>
</div>
{rows.map((row) => row.type === "helper" ? (
<HelperRow key={`helper-${row.step.step_id}`} row={row} turn={turn} total={total} onToggle={() => toggleHelper(row.step.step_id)} />
@@ -459,24 +494,30 @@ export function StepsPanel({
{turn && (modeLabel(turn, t) || stampLine(turn.stamp, t)) && (
<div className="steps-stamp" data-testid="steps-stamp">
{modeLabel(turn, t) && (
- <span className={"steps-pill steps-mode" + (turn.reviewer?.active ? " review" : "")} data-testid="steps-mode">{modeLabel(turn, t)}</span>
+ <span className={"steps-pill steps-mode" + (turn.reviewer?.active ? " review" : "")} data-testid="steps-mode"><Help text={H(t, "mode")}>{modeLabel(turn, t)}</Help></span>
)}
- {stampLine(turn.stamp, t) && <span className="steps-stamp-line">{stampLine(turn.stamp, t)}</span>}
+ {stampLine(turn.stamp, t) && <span className="steps-stamp-line"><Help text={H(t, "stamp")}>{stampLine(turn.stamp, t)}</Help></span>}
</div>
)}
{turn && !inspecting && (
<>
<div className="steps-tiles" data-testid="steps-tiles">
- <Tile value={formatDuration(turn.buckets.total)} label={t("steps.bucket.total")} />
+ <Tile value={formatDuration(turn.buckets.total)} label={t("steps.bucket.total")} help={H(t, "tile.total")} />
{BUCKETS.filter((b) => turn.buckets[b] > 0).map((b) => (
- <Tile key={b} value={formatDuration(turn.buckets[b])} label={t(`steps.bucket.${b}`)} />
+ <Tile key={b} value={formatDuration(turn.buckets[b])} label={t(`steps.bucket.${b}`)} help={H(t, `tile.${b}`)} />
))}
- {turn.buckets.other_shown && <Tile value={formatDuration(turn.buckets.other)} label={t("steps.bucket.other")} />}
+ {turn.buckets.other_shown && <Tile value={formatDuration(turn.buckets.other)} label={t("steps.bucket.other")} help={H(t, "tile.other")} />}
+ {helperCount(steps) > 0 && (
+ <>
+ <Tile value={String(helperCount(steps))} label={t("steps.bucket.helpers")} help={H(t, "tile.helpers")} />
+ <Tile value={formatTokens(helperTokens(steps))} label={t("steps.bucket.helper_tokens")} help={H(t, "tile.helper_tokens")} />
+ </>
+ )}
{gapTotal(steps, "controller_offline") > 0 && (
- <Tile value={formatDuration(gapTotal(steps, "controller_offline"))} label={t("steps.gap.offline_total")} />
+ <Tile value={formatDuration(gapTotal(steps, "controller_offline"))} label={t("steps.gap.offline_total")} help={H(t, "tile.offline")} />
)}
{gapTotal(steps, "nobody_watching") > 0 && (
- <Tile value={formatDuration(gapTotal(steps, "nobody_watching"))} label={t("steps.gap.nobody_total")} />
+ <Tile value={formatDuration(gapTotal(steps, "nobody_watching"))} label={t("steps.gap.nobody_total")} help={H(t, "tile.nobody")} />
)}
</div>
{steps.some((s) => s.status === "unknown") && (
@@ -544,11 +585,11 @@ export function StepsPanel({
);
}
-function Tile({ value, label }: { value: string; label: string }) {
+function Tile({ value, label, help }: { value: string; label: string; help?: string }) {
return (
<div className="steps-tile">
<strong>{value}</strong>
- <span>{label}</span>
+ <span><Help text={help}>{label}</Help></span>
</div>
);
}
@@ -565,6 +606,7 @@ function StepRow({ step, depth, turn, total, label, selected, issue, onSelect }:
<button
className={`steps-row k-${kind}` + (selected ? " selected" : "") + (step.shadow ? " shadow" : "")}
onClick={onSelect}
+ title={kindHelp(step, t) || undefined}
data-testid="steps-row"
>
<span className="steps-label" style={{ paddingLeft: depth * 16 }}>
@@ -599,7 +641,7 @@ function HelperRow({ row, turn, total, onToggle }: {
className="steps-row k-tool steps-helper"
onClick={onToggle}
aria-expanded={row.open}
- title={row.open ? t("steps.helper.hide") : t("steps.helper.show")}
+ title={`${H(t, "kind.helper")} ${row.open ? t("steps.helper.hide") : t("steps.helper.show")}`}
data-testid="steps-helper"
>
<span className="steps-label" style={{ paddingLeft: row.depth * 16 }}>
@@ -648,12 +690,54 @@ function issueTitle(i: { check: string; facts?: Record<string, any> }, t: T): st
: t(`steps.issues.title.${i.check}`, { defaultValue: i.check });
}
+/** An issue's sentence in the reader's language, from the facts it was stored with (TD2-3,
+ * fix list C-2). Turns checked before the facts were stored keep their English sentence. */
+export function issueSentence(i: { check: string; text: string; facts?: Record<string, any> }, t: T): string {
+ const { count, ...f } = i.facts || {};
+ if (!f.words) return i.text;
+ const p: Record<string, any> = {
+ ...f, n: count,
+ extra: typeof f.extra_tokens === "number" ? f.extra_tokens.toLocaleString() : f.extra_tokens,
+ exit: f.exit_code,
+ error: f.error_type ? t(`steps.error_type.${f.error_type}`, { defaultValue: f.error_type }) : "",
+ purpose: f.purpose ? t(`steps.purpose.${f.purpose}`, { defaultValue: f.purpose }) : "",
+ result: f.result ? t(`steps.result.${f.result}`, { defaultValue: f.result }) : "",
+ stage: f.stage ? t(`steps.sbx_stage.${f.stage}`, { defaultValue: f.stage }).toLowerCase() : "",
+ };
+ const variant: Record<string, string> = {
+ off_plan: `off_plan_${f.kind}`, refused: `refused_${f.by}`, made_up_tool: `made_up_tool_${f.kind}`,
+ context_full: `context_full_${f.kind}`,
+ tool_failed: f.exit_code != null ? "tool_failed_exit" : "tool_failed",
+ blocked_by_sandbox: f.exit_code != null ? "blocked_by_sandbox_exit" : "blocked_by_sandbox",
+ waited_unwatched: f.inbox ? "waited_unwatched_inbox" : "waited_unwatched",
+ };
+ let out = t(`steps.issues.say.${variant[i.check] || i.check}`, { ...p, defaultValue: "" });
+ if (!out) return i.text;
+ if (i.check === "tool_failed" && f.recovered) out += " " + t("steps.issues.say.tool_failed_recovered", p);
+ if (i.check === "model_failed" && f.stopped) out += " " + t("steps.issues.say.model_failed_stopped");
+ return out;
+}
+
+/** A "Checked and fine" line in the reader's language, from its facts (C-2). */
+export function fineSentence(f: { check: string; text: string; facts?: Record<string, any> }, t: T): string {
+ const { count, ...x } = f.facts || {};
+ if (!x.words) return f.text;
+ const key = f.check === "reviewer_could_not_run" ? `reviewer_could_not_run_${x.variant}`
+ : f.check === "step_limit" && x.cap == null ? "step_limit_nocap" : f.check;
+ // The reviewer lines say "1 action" or "2 actions": the count picks the plural form.
+ const plural = f.check === "reviewer_could_not_run" ? { count } : {};
+ let out = t(`steps.issues.fine_say.${key}`, { ...x, n: count, ...plural, defaultValue: "" });
+ if (!out) return f.text;
+ if (f.check === "context_full" && x.normal) out += t("steps.issues.fine_say.context_full_normal", { n: x.normal });
+ return out;
+}
+
/** What the app adds from the step's own content, shown here only (design §6): the command a
* failed tool ran and the first line of its output; the call a shadow check judged. */
function IssueDetail({ sessionId, issue }: { sessionId: string; issue: TurnIssue }) {
const [d, setD] = useState<any>(null);
useEffect(() => {
- if (!issue.step_id || !["tool_failed", "shadow_disagreed", "refused", "reviewer_allowed_send", "blocked_by_sandbox"].includes(issue.check)) return;
+ if (!issue.step_id || !["tool_failed", "shadow_disagreed", "refused", "reviewer_allowed_send", "blocked_by_sandbox", "model_failed"].includes(issue.check)) return;
let live = true;
getStepDetail(sessionId, issue.step_id).then((x) => live && setD(x)).catch(() => {});
return () => {
@@ -664,7 +748,10 @@ function IssueDetail({ sessionId, issue }: { sessionId: string; issue: TurnIssue
const args = d.kind === "auto_approver" ? d.for_step?.arguments : d.arguments;
const main = args && typeof args === "object" ? (args.command ?? args.path ?? args.url ?? args.query) : null;
let first = "";
- if (issue.check === "tool_failed" && typeof d.message?.content === "string") {
+ if (issue.check === "model_failed") {
+ // What the provider said (mock review M2-R3), cut to its first line.
+ first = String(d.attrs?.error?.text || "").split("\n").map((l: string) => l.trim()).find(Boolean) || "";
+ } else if (issue.check === "tool_failed" && typeof d.message?.content === "string") {
try {
const r = JSON.parse(d.message.content);
first = String(r?.output ?? r?.error ?? "").split("\n").map((l: string) => l.trim()).find(Boolean) || "";
@@ -695,21 +782,26 @@ function IssuesView({ sessionId, turn, issues, steps, onGo }: {
if (g) g.items.push(i);
else groups.push({ key, items: [i] });
}
- const noPlan = issues.not_checked.find((n) => n.check === "off_plan");
+ const [openNa, setOpenNa] = useState(false);
+ const fired = new Set(issues.issues.map((i) => i.check)).size;
return (
<div className="steps-issues" data-testid="steps-issues">
+ <p className="steps-faint steps-issues-summary" data-testid="steps-issues-summary">
+ <Help text={H(t, "issues.summary")}>{t("steps.issues.summary", {
+ total: fired + issues.fine.length + issues.not_checked.length, fired, fine: issues.fine.length, na: issues.not_checked.length })}</Help>
+ </p>
{groups.length === 0 && <p className="steps-issues-none">{t("steps.issues.none")}</p>}
{groups.map((g) => {
const first = g.items[0];
const info = first.severity === "info";
return (
<div key={g.key} className={"steps-callout " + (info ? "info" : "warn")} data-testid="steps-issue">
- <b>{issueTitle(first, t)}{g.items.length > 1 ? ` · ${g.items.length}` : ""}</b>
- {info && <span className="steps-pill steps-info-pill">{t("steps.issues.information")}</span>}
+ <b><Help text={H(t, `issue.${first.check}`)}>{issueTitle(first, t)}</Help>{g.items.length > 1 ? ` · ${g.items.length}` : ""}</b>
+ {info && <span className="steps-pill steps-info-pill"><Help text={H(t, "issues.information")}>{t("steps.issues.information")}</Help></span>}
{g.items.map((i, n) => (
<div key={(i.step_id || "") + n} className="steps-issue-line">
{i.step_seq != null && <span className="steps-faint">{t("steps.insp.step", { n: i.step_seq })} </span>}
- <span>{i.text}</span>
+ <span>{issueSentence(i, t)}</span>
<IssueDetail sessionId={sessionId} issue={i} />
{i.step_id && steps.some((s) => s.step_id === i.step_id) && (
<button className="steps-link" onClick={() => onGo(i.step_id!)}>{t("steps.issues.go")}</button>
@@ -719,16 +811,28 @@ function IssuesView({ sessionId, turn, issues, steps, onGo }: {
</div>
);
})}
- <button className="steps-fine-toggle" onClick={() => setOpenFine(!openFine)} data-testid="steps-fine-toggle">
+ <button className="steps-fine-toggle" onClick={() => setOpenFine(!openFine)} title={H(t, "issues.fine")} data-testid="steps-fine-toggle">
{t("steps.issues.fine", { n: issues.fine.length })} {openFine ? "▾" : "▸"}
</button>
{openFine && (
<div className="steps-fine" data-testid="steps-fine">
- {issues.fine.map((f) => <div key={f.check}>✓ {f.text}</div>)}
- {noPlan && <div className="steps-faint">– {t("steps.issues.no_plan")}</div>}
+ {issues.fine.map((f) => <div key={f.check}>✓ {fineSentence(f, t)}</div>)}
</div>
)}
- {!openFine && noPlan && <div className="steps-faint steps-fine">– {t("steps.issues.no_plan")}</div>}
+ {issues.not_checked.length > 0 && (
+ <>
+ <button className="steps-fine-toggle" onClick={() => setOpenNa(!openNa)} title={H(t, "issues.na")} data-testid="steps-na-toggle">
+ {t("steps.issues.na", { n: issues.not_checked.length })} {openNa ? "▾" : "▸"}
+ </button>
+ {openNa && (
+ <div className="steps-fine steps-faint" data-testid="steps-na">
+ {issues.not_checked.map((n) => (
+ <div key={n.check}>– {issueTitle(n, t)}: {t(`steps.issues.na_reason.${n.check}`, { defaultValue: n.text || "" })}</div>
+ ))}
+ </div>
+ )}
+ </>
+ )}
</div>
);
}
@@ -754,42 +858,46 @@ function inputLines(request: any): string[] {
});
}
-function Kv({ rows }: { rows: [string, React.ReactNode][] }) {
+type KvRow = [string, React.ReactNode] | [string, React.ReactNode, string];
+
+/** Label and value rows; a third item is the label's hover help (F-6). */
+function Kv({ rows }: { rows: KvRow[] }) {
return (
<dl className="steps-kv">
- {rows.map(([k, v]) => (
- <div key={k}><dt>{k}</dt><dd>{v ?? "—"}</dd></div>
+ {rows.map(([k, v, help]) => (
+ <div key={k}><dt><Help text={help}>{k}</Help></dt><dd>{v ?? "—"}</dd></div>
))}
</dl>
);
}
-function Sec({ label, right }: { label: string; right?: React.ReactNode }) {
- return <div className="steps-seclbl">{label}{right != null && <><span className="steps-sp" /><span className="steps-seclbl-right">{right}</span></>}</div>;
+function Sec({ label, right, help }: { label: string; right?: React.ReactNode; help?: string }) {
+ return <div className="steps-seclbl"><Help text={help}>{label}</Help>{right != null && <><span className="steps-sp" /><span className="steps-seclbl-right">{right}</span></>}</div>;
}
function ContextAdded({ ctx, t }: { ctx: Record<string, any> | null | undefined; t: T }) {
if (!ctx) return null;
const start = ctx.session_start, pt = ctx.per_turn || {};
- const rows: [string, React.ReactNode][] = [];
+ const rows: [string, React.ReactNode, string][] = [];
if (start) {
const mems = start.memories || [];
- rows.push(["memory", <>{t("steps.detail.memories")} {mems.length}{start.memory_mode === "index" ? ` · ${t("steps.detail.as_index")}` : ""}
- {mems.length > 0 && <span className="steps-mono steps-faint"> {mems.slice(0, 6).map((m: any) => `mem_${m}`).join(", ")}</span>}</>]);
- if (start.rules) rows.push(["rules", <>{t("steps.detail.rules")} {start.rules}</>]);
- if (start.agents_md) rows.push(["AGENTS.md", <>{t("steps.detail.agents_md")} {start.agents_md}</>]);
+ // F-2: the count and the mode say what matters; memory ids carried no names.
+ rows.push(["memory", <>{t("steps.detail.memories")} {mems.length}{start.memory_mode === "index" ? ` · ${t("steps.detail.as_index")}` : ""}</>, "ctx.memory"]);
+ // F-5: shown at 0 too, like the other rows, so a missing row never means "not recorded".
+ rows.push(["rules", <>{t("steps.detail.rules")} {start.rules ?? 0}</>, "ctx.rules"]);
+ rows.push(["AGENTS.md", <>{t("steps.detail.agents_md")} {start.agents_md ?? 0}</>, "ctx.agents_md"]);
}
- rows.push(["skill", <>{t("steps.detail.skills")} {pt.skills_menu || 0}{pt.skills_hash && <span className="steps-mono steps-faint"> ⓢ{pt.skills_hash}</span>}</>]);
- rows.push(["folder", <>{t("steps.detail.folders")} {pt.folders ?? 0}</>]);
- if (pt.mode_note) rows.push(["mode", <>{t("steps.detail.mode_note")} {pt.mode_note}</>]);
- if (pt.sandbox_note) rows.push(["sandbox", <>{t("steps.detail.sandbox_note")}</>]);
+ rows.push(["skill", <>{t("steps.detail.skills")} {pt.skills_menu || 0}{pt.skills_hash && <span className="steps-mono steps-faint"> ⓢ{pt.skills_hash}</span>}</>, "ctx.skill"]);
+ rows.push(["folder", <>{t("steps.detail.folders")} {pt.folders ?? 0}</>, "ctx.folder"]);
+ if (pt.mode_note) rows.push(["mode", <>{t("steps.detail.mode_note")} {pt.mode_note}</>, "ctx.mode"]);
+ if (pt.sandbox_note) rows.push(["sandbox", <>{t("steps.detail.sandbox_note")}</>, "ctx.sandbox"]);
return (
<div className="steps-ctxlist">
- {rows.map(([tag, body]) => (
- <div key={tag}><span className="steps-pill">{tag}</span><span className="steps-c">{body}</span></div>
+ {rows.map(([tag, body, help]) => (
+ <div key={tag}><span className="steps-pill">{tag}</span><span className="steps-c"><Help text={H(t, help)}>{body}</Help></span></div>
))}
{ctx.size_chars != null && (
- <div><span className="steps-pill">{t("steps.insp.block")}</span><span className="steps-c">{t("steps.detail.size")}</span>
+ <div><span className="steps-pill">{t("steps.insp.block")}</span><span className="steps-c"><Help text={H(t, "ctx.block")}>{t("steps.detail.size")}</Help></span>
<span className="steps-sz">{ctx.size_chars} · <span className="steps-mono">{ctx.block_hash || "—"}</span></span></div>
)}
</div>
@@ -806,12 +914,12 @@ function TokenBar({ u, t }: { u: any; t: T }) {
<div data-testid="steps-tokens">
<div className="steps-tokbar">{seg(cache, "c-cache")}{seg(written, "c-write")}{seg(fresh, "c-new")}{seg(out, "c-out")}</div>
<div className="steps-legend">
- ■ <span className="c-cache">{t("steps.insp.from_cache", { n: formatTokens(cache) })}</span>
- {" · "}<span className="c-write">{t("steps.insp.written", { n: u.cache_write == null ? "—" : formatTokens(written) })}</span>
- {" · "}<span className="c-new">{t("steps.insp.new", { n: formatTokens(fresh) })}</span>
- {" · "}<span className="c-out">{u.reasoning != null
+ ■ <span className="c-cache"><Help text={H(t, "tok.cache")}>{t("steps.insp.from_cache", { n: formatTokens(cache) })}</Help></span>
+ {" · "}<span className="c-write"><Help text={H(t, "tok.written")}>{t("steps.insp.written", { n: u.cache_write == null ? "—" : formatTokens(written) })}</Help></span>
+ {" · "}<span className="c-new"><Help text={H(t, "tok.new")}>{t("steps.insp.new", { n: formatTokens(fresh) })}</Help></span>
+ {" · "}<span className="c-out"><Help text={H(t, "tok.out")}>{u.reasoning != null
? t("steps.insp.out_reasoning", { n: formatTokens(out), r: formatTokens(u.reasoning) })
- : t("steps.insp.out", { n: formatTokens(out) })}</span>
+ : t("steps.insp.out", { n: formatTokens(out) })}</Help></span>
</div>
</div>
);
@@ -819,11 +927,13 @@ function TokenBar({ u, t }: { u: any; t: T }) {
function Joined({ audit, t }: { audit: any[]; t: T }) {
if (!audit?.length) return null;
- const stages = Array.from(new Set(audit.map((r) => r.stage).filter(Boolean)));
+ const ORDER = ["proposed", "started", "finished"];
+ const stages = Array.from(new Set(audit.map((r) => r.stage).filter(Boolean)))
+ .sort((a, b) => (ORDER.indexOf(a) + 1 || 99) - (ORDER.indexOf(b) + 1 || 99));
return (
<>
- <Sec label={t("steps.insp.joined")} />
- <div className="steps-small">{t("steps.insp.joined_line", { count: audit.length })} · <span className="steps-mono">{stages.join(", ")}</span> · {t("steps.insp.same_call")}</div>
+ <Sec label={t("steps.insp.joined")} help={H(t, "sec.joined")} />
+ <div className="steps-small" data-testid="steps-audit">{t("steps.insp.joined_line", { count: audit.length, stages: stages.join(", ") })}</div>
</>
);
}
@@ -865,7 +975,7 @@ function Inspector({ d, turn, steps, sessionId, t }: { d: any; turn: RecordedTur
const exchanges = (d.exchange_rebuilt || []) as any[];
const exact = exchanges.length > 0 && (
<details className="steps-exchange">
- <summary>{t("steps.exchange", { count: exchanges.length })}</summary>
+ <summary title={H(t, "sec.exact")}>{t("steps.exchange", { count: exchanges.length })}</summary>
{exchanges.map((x: any, i: number) => (
<pre className="steps-box" key={i}>{JSON.stringify(x.request, null, 2)}{"\n\n"}{x.response}</pre>
))}
@@ -887,41 +997,41 @@ function Inspector({ d, turn, steps, sessionId, t }: { d: any; turn: RecordedTur
const lines = inputLines(exchanges.find((x) => !x.auxiliary)?.request);
return (
<>
- <h3 className="steps-h3"><span className={`steps-kind k-${kind}`} />{title}<span className="steps-pill">{t("steps.insp.step", { n: d.seq ?? "—" })}</span></h3>
+ <h3 className="steps-h3"><span className={`steps-kind k-${kind}`} /><Help text={kindHelp(d, t)}>{title}</Help><span className="steps-pill">{t("steps.insp.step", { n: d.seq ?? "—" })}</span></h3>
<div className="steps-meta">
{[a.provider_title || a.provider ? t("steps.insp.via", { model: a.model, provider: a.provider_title || a.provider }) : a.model, t("steps.insp.started", { at: started }), why].filter(Boolean).join(" · ")}
</div>
<Kv rows={[
- [t("steps.detail.duration"), <>{formatDuration(d.duration_ms)}{a.ttft_ms != null && <span className="steps-faint"> ({t("steps.insp.first_reply", { t: formatDuration(a.ttft_ms) })})</span>}</>],
- [t("steps.detail.attempts"), attempts <= 1 ? t("steps.insp.attempt_one") : t("steps.insp.attempt_many", { n: attempts, retries: attempts - 1 })],
+ [t("steps.detail.duration"), <>{formatDuration(d.duration_ms)}{a.ttft_ms != null && <span className="steps-faint"> ({t("steps.insp.first_reply", { t: formatDuration(a.ttft_ms) })})</span>}</>, H(t, "detail.duration")],
+ [t("steps.detail.attempts"), attempts <= 1 ? t("steps.insp.attempt_one") : t("steps.insp.attempt_many", { n: attempts, retries: attempts - 1 }), H(t, "detail.attempts")],
...(facts.served_model && facts.served_model !== String(a.model || "").split(":").pop()
- ? [[t("steps.detail.served"), facts.served_model] as [string, React.ReactNode]] : []),
+ ? [[t("steps.detail.served"), facts.served_model, H(t, "detail.served")] as KvRow] : []),
...(ctxStep?.attrs?.session_prompt_hash
- ? [[t("steps.detail.session_prompt"), <><span className="steps-mono">{ctxStep.attrs.session_prompt_hash}</span> <span className="steps-faint">{t("steps.insp.session_prompt_note")}</span></>] as [string, React.ReactNode]] : []),
+ ? [[t("steps.detail.session_prompt"), <><span className="steps-mono">{ctxStep.attrs.session_prompt_hash}</span> <span className="steps-faint">{t("steps.insp.session_prompt_note")}</span></>, H(t, "detail.session_prompt")] as KvRow] : []),
...(turn.stamp?.persona_fp
- ? [[t("steps.detail.coworker"), <><span className="steps-mono">ⓕ{turn.stamp.persona_fp}</span> <span className="steps-faint">{t("steps.insp.coworker_note")}</span></>] as [string, React.ReactNode]] : []),
- [t("steps.detail.tools"), a.tools_offered ? <>{t("steps.insp.tools_n", { count: a.tools_offered })} <span className="steps-mono steps-faint">set·{a.tools_hash || "—"}</span></> : "0"],
- ...(facts.provider_request_id ? [[t("steps.detail.request_id"), <span className="steps-mono">{facts.provider_request_id}</span>] as [string, React.ReactNode]] : []),
- ...(facts.provider_response_id ? [[t("steps.detail.response_id"), <span className="steps-mono">{facts.provider_response_id}</span>] as [string, React.ReactNode]] : []),
- ...(a.error ? [[t("steps.detail.error"), `${t(`steps.error_type.${a.error.type}`, { defaultValue: a.error.type })}${a.error.text ? ` · ${a.error.text}` : ""}`] as [string, React.ReactNode]] : []),
+ ? [[t("steps.detail.coworker"), <><span className="steps-mono">ⓕ{turn.stamp.persona_fp}</span> <span className="steps-faint">{t("steps.insp.coworker_note")}</span></>, H(t, "detail.coworker")] as KvRow] : []),
+ [t("steps.detail.tools"), a.tools_offered ? <>{t("steps.insp.tools_n", { count: a.tools_offered })} <span className="steps-mono steps-faint">set·{a.tools_hash || "—"}</span></> : "0", H(t, "detail.tools")],
+ ...(facts.provider_request_id ? [[t("steps.detail.request_id"), <span className="steps-mono">{facts.provider_request_id}</span>, H(t, "detail.request_id")] as KvRow] : []),
+ ...(facts.provider_response_id ? [[t("steps.detail.response_id"), <span className="steps-mono">{facts.provider_response_id}</span>, H(t, "detail.response_id")] as KvRow] : []),
+ ...(a.error ? [[t("steps.detail.error"), `${t(`steps.error_type.${a.error.type}`, { defaultValue: a.error.type })}${a.error.text ? ` · ${a.error.text}` : ""}`, H(t, "detail.error")] as KvRow] : []),
]} />
{u && (
<>
- <Sec label={t("steps.insp.tokens")} right={u.reported ? t("steps.insp.in_out", {
+ <Sec label={t("steps.insp.tokens")} help={H(t, "sec.tokens")} right={u.reported ? t("steps.insp.in_out", {
in: formatTokens((u.cache_read || 0) + (u.cache_write || 0) + (u.input_fresh || 0)), out: formatTokens(u.output || 0) }) : undefined} />
<TokenBar u={u} t={t} />
</>
)}
{(a.purpose || "main") === "main" && ctxStep && (
<>
- <Sec label={t("steps.insp.context_added")} />
+ <Sec label={t("steps.insp.context_added")} help={H(t, "sec.context_added")} />
<ContextAdded ctx={ctxStep.attrs} t={t} />
</>
)}
{lines.length > 0 && <InputBox lines={lines} t={t} />}
{(d.message?.content || d.message?.tool_calls?.length) && (
<>
- <Sec label={t("steps.insp.output")} />
+ <Sec label={t("steps.insp.output")} help={H(t, "sec.output")} />
<div className="steps-box">
{[d.message.content ? `${d.message.content}${d.message.cut ? "\n…" : ""}` : "",
...(d.message.tool_calls || []).map((c: any) => `→ ${t("steps.insp.tool_call")}: ${c.name} ${c.arguments}`)].filter(Boolean).join("\n")}
@@ -948,23 +1058,23 @@ function Inspector({ d, turn, steps, sessionId, t }: { d: any; turn: RecordedTur
const who = turn.reviewer?.active ? reviewerLabel(turn.reviewer, t) : "";
return (
<>
- <h3 className="steps-h3"><span className={`steps-kind k-${kind}`} />{d.name}<span className={`steps-pill ${pill.cls}`}>{pill.text}</span></h3>
+ <h3 className="steps-h3"><span className={`steps-kind k-${kind}`} /><Help text={kindHelp(d, t)}>{d.name}</Help><span className={`steps-pill ${pill.cls}`}>{pill.text}</span></h3>
<div className="steps-meta">
{[a.risk_class ? t(`steps.risk.${a.risk_class}`, { defaultValue: a.risk_class }) : "",
d.call_id ? `${t("steps.detail.call_id")} ${d.call_id}` : "",
mode ? `${t("steps.detail.mode")} ${mode}${who ? ` (${t("steps.reviewer_is", { who })})` : ""}` : ""].filter(Boolean).join(" · ")}
</div>
<Kv rows={[
- [t("steps.insp.ran_for"), formatDuration(d.duration_ms)],
- ...(wait ? [[t("steps.insp.waited"), formatDuration(wait.duration_ms)] as [string, React.ReactNode]] : []),
- ...(a.result_bytes != null ? [[t("steps.detail.result_size"), formatBytes(a.result_bytes)] as [string, React.ReactNode]] : []),
- ...(a.exit_code != null ? [[t("steps.detail.exit"), String(a.exit_code)] as [string, React.ReactNode]] : []),
- ...(a.decision?.origin ? [[t("steps.detail.origin"), t(`steps.origin.${a.decision.origin}`, { defaultValue: a.decision.origin })] as [string, React.ReactNode]] : []),
- ...(a.sandbox ? [[t("steps.detail.blocked"), a.sandbox.blocked_known ? String(a.sandbox.blocked_count) : t("steps.detail.unknown")] as [string, React.ReactNode]] : []),
+ [t("steps.insp.ran_for"), formatDuration(d.duration_ms), H(t, "detail.ran_for")],
+ ...(wait ? [[t("steps.insp.waited"), formatDuration(wait.duration_ms), H(t, "detail.waited")] as KvRow] : []),
+ ...(a.result_bytes != null ? [[t("steps.detail.result_size"), formatBytes(a.result_bytes), H(t, "detail.result_size")] as KvRow] : []),
+ ...(a.exit_code != null ? [[t("steps.detail.exit"), String(a.exit_code), H(t, "detail.exit")] as KvRow] : []),
+ ...(a.decision?.origin ? [[t("steps.detail.origin"), t(`steps.origin.${a.decision.origin}`, { defaultValue: a.decision.origin }), H(t, "detail.origin")] as KvRow] : []),
+ ...(a.sandbox ? [[t("steps.detail.blocked"), a.sandbox.blocked_known ? String(a.sandbox.blocked_count) : t("steps.detail.unknown"), H(t, "detail.blocked")] as KvRow] : []),
]} />
{path.length > 0 && (
<>
- <Sec label={t("steps.insp.how_allowed")} />
+ <Sec label={t("steps.insp.how_allowed")} help={H(t, "sec.how_allowed")} />
<ol className="steps-ctxlist steps-path">
{path.map((p: any, i: number) => {
const last = i === path.length - 1;
@@ -983,13 +1093,13 @@ function Inspector({ d, turn, steps, sessionId, t }: { d: any; turn: RecordedTur
)}
{d.arguments != null && (
<>
- <Sec label={t("steps.insp.arguments")} />
+ <Sec label={t("steps.insp.arguments")} help={H(t, "sec.arguments")} />
<div className="steps-box">{typeof d.arguments === "string" ? d.arguments : JSON.stringify(d.arguments, null, 2)}</div>
</>
)}
{d.message?.content && (
<>
- <Sec label={t("steps.insp.result")} />
+ <Sec label={t("steps.insp.result")} help={H(t, "sec.result")} />
<div className="steps-box">{d.message.content}{d.message.cut ? "\n…" : ""}</div>
</>
)}
@@ -1002,46 +1112,46 @@ function Inspector({ d, turn, steps, sessionId, t }: { d: any; turn: RecordedTur
// Waits, summaries, context steps and notices.
const pill = statusPill(d, t);
- const rows: [string, React.ReactNode][] = [
- [t("steps.detail.started"), started],
- [t("steps.detail.duration"), formatDuration(d.duration_ms)],
+ const rows: KvRow[] = [
+ [t("steps.detail.started"), started, H(t, "detail.started")],
+ [t("steps.detail.duration"), formatDuration(d.duration_ms), H(t, "detail.duration")],
];
if (d.kind === "wait") {
- rows.push([t("steps.detail.where"), t(`steps.where.${a.where}`, { defaultValue: a.where })],
- [t("steps.detail.outcome"), a.outcome == null ? "—" : t(`steps.outcome.${a.outcome}`, { defaultValue: a.outcome })]);
- if (d.content?.resolved_by) rows.push([t("steps.detail.answered_by"), d.content.resolved_by]);
+ rows.push([t("steps.detail.where"), t(`steps.where.${a.where}`, { defaultValue: a.where }), H(t, "detail.where")],
+ [t("steps.detail.outcome"), a.outcome == null ? "—" : t(`steps.outcome.${a.outcome}`, { defaultValue: a.outcome }), H(t, "detail.outcome")]);
+ if (d.content?.resolved_by) rows.push([t("steps.detail.answered_by"), d.content.resolved_by, H(t, "detail.answered_by")]);
}
if (d.kind === "summary") {
- if (a.reason) rows.push([t("steps.detail.summary_reason"), t(`steps.summary_reason.${a.reason}`, { defaultValue: a.reason })]);
- rows.push([t("steps.detail.tokens_before"), a.tokens_before == null ? "—" : formatTokens(a.tokens_before)],
- [t("steps.detail.tokens_after"), a.tokens_after == null ? "—" : formatTokens(a.tokens_after)]);
+ if (a.reason) rows.push([t("steps.detail.summary_reason"), t(`steps.summary_reason.${a.reason}`, { defaultValue: a.reason }), H(t, "detail.summary_reason")]);
+ rows.push([t("steps.detail.tokens_before"), a.tokens_before == null ? "—" : formatTokens(a.tokens_before), H(t, "detail.tokens_before")],
+ [t("steps.detail.tokens_after"), a.tokens_after == null ? "—" : formatTokens(a.tokens_after), H(t, "detail.tokens_after")]);
}
- if (d.kind === "context") rows.push([t("steps.detail.session_prompt"), <span className="steps-mono">{a.session_prompt_hash || "—"}</span>]);
+ if (d.kind === "context") rows.push([t("steps.detail.session_prompt"), <span className="steps-mono">{a.session_prompt_hash || "—"}</span>, H(t, "detail.session_prompt")]);
if (d.kind === "sandbox") {
- rows.push([t("steps.sbx.provider"), a.provider || "—"],
- [t("steps.sbx.trigger"), t(`steps.sbx.trigger_${a.trigger}`, { defaultValue: a.trigger || "—" })]);
+ rows.push([t("steps.sbx.provider"), a.provider || "—", H(t, "detail.sbx_provider")],
+ [t("steps.sbx.trigger"), t(`steps.sbx.trigger_${a.trigger}`, { defaultValue: a.trigger || "—" }), H(t, "detail.sbx_trigger")]);
}
if (d.kind === "notice" && d.name === "sandbox_restarted") {
- rows.push([t("steps.sbx.provider"), a.provider || "—"],
- [t("steps.sbx.reason"), t(`steps.sbx.reason_${a.reason}`, { defaultValue: a.reason || "—" })],
- [t("steps.sbx.how"), t(`steps.sbx.how_${a.how}`, { defaultValue: a.how || "—" })]);
+ rows.push([t("steps.sbx.provider"), a.provider || "—", H(t, "detail.sbx_provider")],
+ [t("steps.sbx.reason"), t(`steps.sbx.reason_${a.reason}`, { defaultValue: a.reason || "—" }), H(t, "detail.sbx_reason")],
+ [t("steps.sbx.how"), t(`steps.sbx.how_${a.how}`, { defaultValue: a.how || "—" }), H(t, "detail.sbx_how")]);
}
if (isGap(d) || (d.kind === "notice" && d.name === "recording_gap")) {
const until = d.name === "recording_gap" ? a.until : d.started_at + (d.duration_ms || 0);
rows.splice(0, rows.length,
- [t("steps.gap.from"), clockTime(d.started_at)],
- [t("steps.gap.until"), typeof until === "number" ? clockTime(until) : "—"]);
- if (d.name === "clock_jump") rows.push([t("steps.gap.skipped"), formatDuration(a.skipped_ms ?? d.duration_ms)]);
- else rows.push([t("steps.detail.duration"), formatDuration(typeof until === "number" ? until - d.started_at : d.duration_ms)]);
- if (a.reason) rows.push([t("steps.gap.why"), t(`steps.gap.reason_${a.reason}`, { defaultValue: a.reason })]);
- if (a.cause) rows.push([t("steps.gap.why"), t(`steps.gap.cause_${a.cause}`, { defaultValue: a.cause })]);
+ [t("steps.gap.from"), clockTime(d.started_at), H(t, "detail.gap_from")],
+ [t("steps.gap.until"), typeof until === "number" ? clockTime(until) : "—", H(t, "detail.gap_until")]);
+ if (d.name === "clock_jump") rows.push([t("steps.gap.skipped"), formatDuration(a.skipped_ms ?? d.duration_ms), H(t, "detail.gap_skipped")]);
+ else rows.push([t("steps.detail.duration"), formatDuration(typeof until === "number" ? until - d.started_at : d.duration_ms), H(t, "detail.duration")]);
+ if (a.reason) rows.push([t("steps.gap.why"), t(`steps.gap.reason_${a.reason}`, { defaultValue: a.reason }), H(t, "detail.gap_why")]);
+ if (a.cause) rows.push([t("steps.gap.why"), t(`steps.gap.cause_${a.cause}`, { defaultValue: a.cause }), H(t, "detail.gap_why")]);
}
const sbxStages: SandboxStage[] = d.kind === "sandbox" && Array.isArray(a.stages) ? a.stages : [];
const changed = d.kind === "notice" && d.name === "stamp_changed" ? stampChanges(a, t) : [];
const showsContext = d.kind === "context" || (d.kind === "notice" && d.name === "context_changed");
return (
<>
- <h3 className="steps-h3"><span className={`steps-kind k-${kind}`} />{stepLabel(d, t)}<span className={`steps-pill ${pill.cls}`}>{pill.text}</span></h3>
+ <h3 className="steps-h3"><span className={`steps-kind k-${kind}`} /><Help text={kindHelp(d, t)}>{stepLabel(d, t)}</Help><span className={`steps-pill ${pill.cls}`}>{pill.text}</span></h3>
{d.kind === "sandbox" && d.status === "error" && (
<div className="steps-callout err" data-testid="steps-sandbox-failed">
{t("steps.sbx.failed", {
@@ -1053,35 +1163,36 @@ function Inspector({ d, turn, steps, sessionId, t }: { d: any; turn: RecordedTur
<Kv rows={rows} />
{sbxStages.length > 0 && (
<>
- <Sec label={t("steps.sbx.stages")} />
+ <Sec label={t("steps.sbx.stages")} help={H(t, "sec.stages")} />
<Kv rows={sbxStages.map((s) => [
t(`steps.sbx_stage.${s.name}`, { defaultValue: s.name }),
s.uncertain ? t("steps.sbx.image_note") : `${formatDuration(s.duration_ms)}${s.status === "error" ? ` · ${t("steps.status.error")}` : ""}`,
- ] as [string, React.ReactNode])} />
+ ] as KvRow)} />
</>
)}
{d.kind === "notice" && d.name === "clock_jump" && <p className="steps-line">{t("steps.gap.asleep_note")}</p>}
{d.kind === "sandbox" && a.error?.text && (
<>
- <Sec label={t("steps.sbx.error_text")} />
+ <Sec label={t("steps.sbx.error_text")} help={H(t, "sec.error_text")} />
<div className="steps-box">{a.error.text}</div>
</>
)}
{changed.length > 0 && (
<>
- <Sec label={t("steps.insp.changed")} />
+ <Sec label={t("steps.insp.changed")} help={H(t, "sec.changed")} />
<Kv rows={changed} />
</>
)}
{showsContext && (
<>
- <Sec label={d.kind === "context" ? t("steps.insp.context_added") : t("steps.insp.context_now")} />
+ <Sec label={d.kind === "context" ? t("steps.insp.context_added") : t("steps.insp.context_now")}
+ help={H(t, d.kind === "context" ? "sec.context_added" : "sec.context_now")} />
<ContextAdded ctx={a} t={t} />
</>
)}
{d.children?.length > 0 && (
<>
- <Sec label={t("steps.insp.inside")} />
+ <Sec label={t("steps.insp.inside")} help={H(t, "sec.inside")} />
<div className="steps-ctxlist">
{d.children.map((c: any) => (
<div key={c.step_id}><span className={`steps-kind k-${KIND_CLASS[c.kind] || "ctx"}`} /><span className="steps-c">{stepLabel(c, t)}</span><span className="steps-sz">{formatDuration(c.duration_ms)}</span></div>
@@ -1096,8 +1207,8 @@ function Inspector({ d, turn, steps, sessionId, t }: { d: any; turn: RecordedTur
/** A `stamp_changed` notice's new values: only the keys that changed (attended, reviewer,
* the sandbox group). */
-function stampChanges(a: Record<string, any>, t: T): [string, React.ReactNode][] {
- const out: [string, React.ReactNode][] = [];
+function stampChanges(a: Record<string, any>, t: T): KvRow[] {
+ const out: KvRow[] = [];
if (typeof a.attended === "boolean")
out.push([t("steps.insp.changed_attended"), t(a.attended ? "steps.insp.attended_yes" : "steps.insp.attended_no")]);
if (a.reviewer) {
@@ -1123,6 +1234,7 @@ function InputBox({ lines, t }: { lines: string[]; t: T }) {
<>
<Sec
label={all ? t("steps.insp.input_all", { total: lines.length }) : t("steps.insp.input_last", { n: shown.length, total: lines.length })}
+ help={H(t, "sec.input")}
right={lines.length > 3 ? <button className="steps-btn" onClick={() => setAll(!all)}>{all ? t("steps.insp.show_less") : t("steps.insp.show_all")}</button> : undefined}
/>
<div className={"steps-box" + (all ? " tall" : "")}>{shown.join("\n\n")}</div>
@@ -1164,7 +1276,7 @@ function CheckInspector({ d, turn, sessionId, t }: { d: any; turn: RecordedTurn;
<>
<h3 className="steps-h3">
<span className="steps-kind k-review" />
- {d.shadow ? t("steps.insp.shadow_title") : t("steps.insp.check_title")}
+ <Help text={kindHelp(d, t)}>{d.shadow ? t("steps.insp.shadow_title") : t("steps.insp.check_title")}</Help>
<span className={`steps-pill ${d.shadow ? "" : pillCls}`}>{d.shadow ? t("steps.check_mode.shadow") : t(`steps.result.${result}`, { defaultValue: result || "" })}</span>
</h3>
<div className="steps-meta">
@@ -1172,23 +1284,23 @@ function CheckInspector({ d, turn, sessionId, t }: { d: any; turn: RecordedTurn;
</div>
{result === "could_not_run" && <div className="steps-callout err">{t("steps.insp.cnr")}</div>}
<Kv rows={d.shadow ? [
- [t("steps.insp.would"), t(`steps.result.${result}`, { defaultValue: result || "" })],
- [t("steps.insp.you_decided"), wait ? <><b>{t(`steps.outcome.${wait.attrs?.outcome}`, { defaultValue: wait.attrs?.outcome || "—" })}</b> {t("steps.insp.after", { t: formatDuration(wait.duration_ms) })}</> : "—"],
- [t("steps.insp.match"), a.matched_user == null ? "—" : <span className={"steps-pill " + (a.matched_user ? "ok" : "warn")}>{a.matched_user ? t("steps.insp.matched") : t("steps.insp.not_matched")}</span>],
- [t("steps.insp.took"), formatDuration(d.duration_ms)],
- [t("steps.insp.tokens"), tokens],
+ [t("steps.insp.would"), t(`steps.result.${result}`, { defaultValue: result || "" }), H(t, "detail.would")],
+ [t("steps.insp.you_decided"), wait ? <><b>{t(`steps.outcome.${wait.attrs?.outcome}`, { defaultValue: wait.attrs?.outcome || "—" })}</b> {t("steps.insp.after", { t: formatDuration(wait.duration_ms) })}</> : "—", H(t, "detail.you_decided")],
+ [t("steps.insp.match"), a.matched_user == null ? "—" : <span className={"steps-pill " + (a.matched_user ? "ok" : "warn")}>{a.matched_user ? t("steps.insp.matched") : t("steps.insp.not_matched")}</span>, H(t, "detail.match")],
+ [t("steps.insp.took"), formatDuration(d.duration_ms), H(t, "detail.took")],
+ [t("steps.insp.tokens"), tokens, H(t, "detail.check_tokens")],
] : [
- [t("steps.insp.result_label"), t(`steps.insp.result_${result}`, { defaultValue: t(`steps.result.${result}`, { defaultValue: result || "—" }) })],
- ...(d.content?.reason ? [[t("steps.detail.reason"), `"${d.content.reason}"`] as [string, React.ReactNode]] : []),
- [t("steps.insp.took"), formatDuration(d.duration_ms)],
- [t("steps.insp.tokens"), tokens],
+ [t("steps.insp.result_label"), t(`steps.insp.result_${result}`, { defaultValue: t(`steps.result.${result}`, { defaultValue: result || "—" }) }), H(t, "detail.result")],
+ ...(d.content?.reason ? [[t("steps.detail.reason"), `"${d.content.reason}"`, H(t, "detail.reason")] as KvRow] : []),
+ [t("steps.insp.took"), formatDuration(d.duration_ms), H(t, "detail.took")],
+ [t("steps.insp.tokens"), tokens, H(t, "detail.check_tokens")],
...(a.breaker ? [[t("steps.detail.breaker"), a.breaker.tripped
? t("steps.insp.breaker_paused", { n: a.breaker.denials_in_row })
- : t("steps.insp.breaker_armed", { n: a.breaker.denials_in_row ?? 0 })] as [string, React.ReactNode]] : []),
+ : t("steps.insp.breaker_armed", { n: a.breaker.denials_in_row ?? 0 }), H(t, "detail.breaker")] as KvRow] : []),
]} />
{questions.length > 0 && (
<>
- <Sec label={t("steps.insp.questions")} />
+ <Sec label={t("steps.insp.questions")} help={H(t, "sec.questions")} />
<table className="steps-qtable" data-testid="steps-questions">
<thead><tr><th>{t("steps.insp.question")}</th><th>{t("steps.insp.answer")}</th></tr></thead>
<tbody>
@@ -1202,7 +1314,7 @@ function CheckInspector({ d, turn, sessionId, t }: { d: any; turn: RecordedTurn;
)}
{shown && (
<>
- <Sec label={t("steps.insp.shown")} />
+ <Sec label={t("steps.insp.shown")} help={H(t, "sec.shown")} />
<div className="steps-box">{shown}</div>
</>
)}
@@ -1233,14 +1345,14 @@ function UsageView({ sessionId, refreshKey }: { sessionId: string; refreshKey: n
<div className="steps-seclbl steps-usage-title">{t("steps.usage.title")}<span className="steps-sp" />
{u.turns != null && <span className="steps-seclbl-right">{t("steps.usage.turns", { count: u.turns })}</span>}</div>
<div className="steps-tiles">
- <Tile value={(u.at_least ? "≥ " : "") + formatTokens(u.totals?.input || 0)} label={t("steps.usage.input")} />
- <Tile value={formatTokens(u.totals?.output || 0)} label={t("steps.usage.output")} />
- <Tile value={u.cache_share == null ? "—" : `${Math.round(u.cache_share * 100)}%`} label={t("steps.usage.from_cache")} />
+ <Tile value={(u.at_least ? "≥ " : "") + formatTokens(u.totals?.input || 0)} label={t("steps.usage.input")} help={H(t, "usage.input")} />
+ <Tile value={formatTokens(u.totals?.output || 0)} label={t("steps.usage.output")} help={H(t, "usage.output")} />
+ <Tile value={u.cache_share == null ? "—" : `${Math.round(u.cache_share * 100)}%`} label={t("steps.usage.from_cache")} help={H(t, "usage.cache")} />
</div>
{u.unreported_calls > 0 && <p className="steps-line">{t("steps.usage.unreported", { count: u.unreported_calls })}</p>}
{u.failed_calls > 0 && <p className="steps-line">{t("steps.usage.failed", { count: u.failed_calls })}</p>}
{u.overflow_calls > 0 && <p className="steps-line">{t("steps.usage.overflow", { count: u.overflow_calls })}</p>}
- <Sec label={t("steps.usage.by_purpose")} />
+ <Sec label={t("steps.usage.by_purpose")} help={H(t, "usage.purpose")} />
<div className="steps-purposes">
{purposes.map(([p, n]) => (
<div className="steps-purpose" key={p}>
@@ -1250,7 +1362,7 @@ function UsageView({ sessionId, refreshKey }: { sessionId: string; refreshKey: n
</div>
))}
</div>
- <Sec label={t("steps.usage.by_model")} />
+ <Sec label={t("steps.usage.by_model")} help={H(t, "usage.model")} />
<table className="steps-models">
<thead><tr><th>{t("steps.usage.model")}</th><th>{t("steps.usage.in")}</th><th>{t("steps.usage.cache")}</th><th>{t("steps.usage.out")}</th></tr></thead>
<tbody>
@@ -1264,13 +1376,13 @@ function UsageView({ sessionId, refreshKey }: { sessionId: string; refreshKey: n
</table>
{u.context && u.context.window ? (
<>
- <Sec label={t("steps.usage.context_now")} />
+ <Sec label={t("steps.usage.context_now")} help={H(t, "usage.context")} />
<div className="steps-purpose-bar wide"><i className="c-new" style={{ width: `${Math.min(100, (u.context.tokens / u.context.window) * 100)}%` }} /></div>
<p className="steps-line">{t("steps.usage.context_line", { used: formatTokens(u.context.tokens), window: formatTokens(u.context.window) })}</p>
</>
) : u.context?.tokens ? (
<>
- <Sec label={t("steps.usage.context_now")} />
+ <Sec label={t("steps.usage.context_now")} help={H(t, "usage.context")} />
<p className="steps-line">{t("steps.usage.context_no_window", { used: formatTokens(u.context.tokens) })}</p>
</>
) : null}
diff --git a/surfaces/gui/src/components/__fixtures__/issues-golden.json b/surfaces/gui/src/components/__fixtures__/issues-golden.json
new file mode 100644
index 00000000..0b13aefc
--- /dev/null
+++ b/surfaces/gui/src/components/__fixtures__/issues-golden.json
@@ -0,0 +1,4124 @@
+[
+ {
+ "scenario": "S2-01",
+ "kind": "issue",
+ "check": "repeated_step",
+ "text": "read_file was called 3 times in a row with the same arguments. That cost 4,400 extra tokens.",
+ "facts": {
+ "words": 1,
+ "count": 3,
+ "extra_tokens": 4400,
+ "name": "read_file"
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (4 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 4,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-01",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (5 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 5,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-01c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "issue",
+ "check": "off_plan",
+ "text": "A plan step was dropped before it was done (1 of 3).",
+ "facts": {
+ "words": 1,
+ "kind": "dropped",
+ "dropped": 1,
+ "of": 3
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (3 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 3,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-02",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "issue",
+ "check": "off_plan",
+ "text": "It kept acting after it marked every plan step done: run_shell.",
+ "facts": {
+ "words": 1,
+ "kind": "after_done",
+ "name": "run_shell"
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (4 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 4,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-02b",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "off_plan",
+ "text": "No plan step was dropped (2 of 2 done)",
+ "facts": {
+ "words": 1,
+ "done": 2,
+ "items": 2
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (4 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 4,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-02c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-02n",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "issue",
+ "check": "refused",
+ "text": "You refused run_shell.",
+ "facts": {
+ "words": 1,
+ "by": "user",
+ "name": "run_shell"
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-03",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (1 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 1,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-03b",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "issue",
+ "check": "overridden",
+ "text": "The auto-approver blocked run_shell and you used Allow anyway.",
+ "facts": {
+ "words": 1,
+ "name": "run_shell"
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "reviewer_could_not_run",
+ "text": "Auto-approver checked 1 action: 0 allowed, 0 asked you, 1 blocked; none failed to run",
+ "facts": {
+ "words": 1,
+ "variant": "live",
+ "count": 1,
+ "allowed": 0,
+ "asked": 0,
+ "blocked": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "reviewer_paused",
+ "text": "Auto-approver was never paused",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "reviewer_allowed_send",
+ "text": "The auto-approver let nothing leave the machine without asking you",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-04",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-04n",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "issue",
+ "check": "reviewer_could_not_run",
+ "text": "The auto-approver could not check an action, so it was shown to you instead.",
+ "facts": {
+ "words": 1,
+ "name": "an action"
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "overridden",
+ "text": "You did not override the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "reviewer_paused",
+ "text": "Auto-approver was never paused",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "reviewer_allowed_send",
+ "text": "The auto-approver let nothing leave the machine without asking you",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-05",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "overridden",
+ "text": "You did not override the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "reviewer_could_not_run",
+ "text": "Auto-approver checked 1 action: 1 allowed, 0 asked you, 0 blocked; none failed to run",
+ "facts": {
+ "words": 1,
+ "variant": "live",
+ "count": 1,
+ "allowed": 1,
+ "asked": 0,
+ "blocked": 0
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "reviewer_paused",
+ "text": "Auto-approver was never paused",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "reviewer_allowed_send",
+ "text": "The auto-approver let nothing leave the machine without asking you",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-05c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "issue",
+ "check": "reviewer_paused",
+ "text": "The auto-approver paused after 3 refusals in a row. Later actions were shown to you.",
+ "facts": {
+ "words": 1,
+ "denials": 3
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "overridden",
+ "text": "You did not override the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "reviewer_could_not_run",
+ "text": "Auto-approver checked 1 action: 0 allowed, 0 asked you, 1 blocked; none failed to run",
+ "facts": {
+ "words": 1,
+ "variant": "live",
+ "count": 1,
+ "allowed": 0,
+ "asked": 0,
+ "blocked": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "reviewer_allowed_send",
+ "text": "The auto-approver let nothing leave the machine without asking you",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-06",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "issue",
+ "check": "shadow_disagreed",
+ "text": "The shadow reviewer would have said allow on an action.",
+ "facts": {
+ "words": 1,
+ "result": "allow",
+ "name": "an action"
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "reviewer_could_not_run",
+ "text": "Shadow reviewer checked 1 action; none failed to run",
+ "facts": {
+ "words": 1,
+ "variant": "shadow",
+ "count": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-07",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "issue",
+ "check": "tool_failed",
+ "text": "run_shell failed with exit code 2.",
+ "facts": {
+ "words": 1,
+ "recovered": false,
+ "name": "run_shell",
+ "exit_code": 2,
+ "how": "exit"
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-08",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "issue",
+ "check": "tool_failed",
+ "text": "read_file failed.",
+ "facts": {
+ "words": 1,
+ "recovered": false,
+ "name": "read_file",
+ "exit_code": null,
+ "how": "result_error"
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-08b",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "issue",
+ "check": "made_up_tool",
+ "text": "The model asked for a tool that does not exist: nope.",
+ "facts": {
+ "words": 1,
+ "kind": "unknown_tool",
+ "name": "nope"
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-08c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "issue",
+ "check": "model_failed",
+ "text": "The model call failed (auth). The turn stopped here.",
+ "facts": {
+ "words": 1,
+ "purpose": "main",
+ "error_type": "auth",
+ "stopped": true
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (1 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 1,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-09",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-09c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "issue",
+ "check": "retried",
+ "text": "The model call needed 2 tries (main).",
+ "facts": {
+ "words": 1,
+ "attempts": 2,
+ "how": "wire",
+ "purpose": "main"
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (1 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 1,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-10",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (1 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 1,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-10b",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "issue",
+ "check": "cut_off",
+ "text": "The model stopped at its output limit. The answer may be incomplete.",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (1 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 1,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-11",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "issue",
+ "check": "step_limit",
+ "text": "The turn stopped after 2 steps, the limit.",
+ "facts": {
+ "words": 1,
+ "iterations": 2
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-12",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "issue",
+ "check": "made_up_tool",
+ "text": "The model asked for a tool that does not exist: search_web_pages.",
+ "facts": {
+ "words": 1,
+ "kind": "unknown_tool",
+ "name": "search_web_pages"
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "issue",
+ "check": "made_up_tool",
+ "text": "The model sent arguments that do not fit write_file.",
+ "facts": {
+ "words": 1,
+ "kind": "bad_arguments",
+ "name": "write_file"
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-13",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "issue",
+ "check": "context_full",
+ "text": "The conversation became too long for the model, so it was summarised in a hurry.",
+ "facts": {
+ "words": 1,
+ "kind": "overflow"
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 79% of the window); summarised 1 time(s) at the normal point",
+ "facts": {
+ "words": 1,
+ "pct": 79,
+ "normal": 1
+ }
+ },
+ {
+ "scenario": "S2-14b",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "issue",
+ "check": "context_full",
+ "text": "The context was 96% full on this call.",
+ "facts": {
+ "words": 1,
+ "kind": "hard_fill",
+ "fill": 0.96,
+ "pct": 96
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (1 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 1,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-14d",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "issue",
+ "check": "blocked_by_sandbox",
+ "text": "The sandbox blocked 2 network request(s) during run_shell, which exited with code 6.",
+ "facts": {
+ "words": 1,
+ "count": 2,
+ "name": "run_shell",
+ "exit_code": 6
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-15",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "blocked_by_sandbox",
+ "text": "The sandbox blocked nothing",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-15c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "issue",
+ "check": "ran_without_sandbox",
+ "text": "This machine has a sandbox switched on, but this turn ran 1 action(s) without it.",
+ "facts": {
+ "words": 1,
+ "count": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-16",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "ran_without_sandbox",
+ "text": "Ran inside the sandbox",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-16c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "issue",
+ "check": "waited_unwatched",
+ "text": "waited 20 min in the Inbox for an answer while nobody was at the app.",
+ "facts": {
+ "words": 1,
+ "minutes": 20,
+ "inbox": true
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-17b",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-17c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "issue",
+ "check": "waited_unwatched",
+ "text": "waited 18 min for an answer while nobody was at the app.",
+ "facts": {
+ "words": 1,
+ "minutes": 18,
+ "inbox": false
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-17d",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "issue",
+ "check": "sandbox_failed",
+ "text": "The sandbox could not start at create: OpenShellImageMissing.",
+ "facts": {
+ "words": 1,
+ "stage": "create",
+ "type": "OpenShellImageMissing"
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (0 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 0,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "blocked_by_sandbox",
+ "text": "The sandbox blocked nothing",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "blocked_by_sandbox",
+ "text": "The sandbox blocked nothing",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "sandbox_failed",
+ "text": "The sandbox started (openshell, 41 s)",
+ "facts": {
+ "words": 1,
+ "provider": "openshell",
+ "seconds": 41
+ }
+ },
+ {
+ "scenario": "S2-18c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "issue",
+ "check": "reviewer_allowed_send",
+ "text": "The auto-approver allowed run_shell without asking you. It sends data off the machine.",
+ "facts": {
+ "words": 1,
+ "name": "run_shell"
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "overridden",
+ "text": "You did not override the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "reviewer_could_not_run",
+ "text": "Auto-approver checked 1 action: 1 allowed, 0 asked you, 0 blocked; none failed to run",
+ "facts": {
+ "words": 1,
+ "variant": "live",
+ "count": 1,
+ "allowed": 1,
+ "asked": 0,
+ "blocked": 0
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "reviewer_paused",
+ "text": "Auto-approver was never paused",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-19",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "repeated_step",
+ "text": "No step repeated 3 times in a row",
+ "facts": {
+ "words": 1,
+ "min_run": 3
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "refused",
+ "text": "No actions were refused by you, a rule or the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "overridden",
+ "text": "You did not override the auto-approver",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "reviewer_could_not_run",
+ "text": "Auto-approver checked 1 action: 1 allowed, 0 asked you, 0 blocked; none failed to run",
+ "facts": {
+ "words": 1,
+ "variant": "live",
+ "count": 1,
+ "allowed": 1,
+ "asked": 0,
+ "blocked": 0
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "reviewer_paused",
+ "text": "Auto-approver was never paused",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "reviewer_allowed_send",
+ "text": "The auto-approver let nothing leave the machine without asking you",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "tool_failed",
+ "text": "No tool failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "model_failed",
+ "text": "No model call failed",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "retried",
+ "text": "No model call needed a second try",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "cut_off",
+ "text": "No answers were cut off (length limit)",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "step_limit",
+ "text": "Stayed under the step limit (2 of 50)",
+ "facts": {
+ "words": 1,
+ "iterations": 2,
+ "cap": 50
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "made_up_tool",
+ "text": "No unknown tools or invalid arguments",
+ "facts": {
+ "words": 1
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "context_full",
+ "text": "Context had room (at most 1% of the window)",
+ "facts": {
+ "words": 1,
+ "pct": 1,
+ "normal": 0
+ }
+ },
+ {
+ "scenario": "S2-19c",
+ "kind": "fine",
+ "check": "waited_unwatched",
+ "text": "Nothing waited long for an answer while nobody was watching",
+ "facts": {
+ "words": 1
+ }
+ }
+]
\ No newline at end of file
diff --git a/surfaces/gui/src/locales/en.json b/surfaces/gui/src/locales/en.json
index 7acc799d..a33d6219 100644
--- a/surfaces/gui/src/locales/en.json
+++ b/surfaces/gui/src/locales/en.json
@@ -3037,6 +3037,172 @@
}
},
"steps": {
+ "help": {
+ "mode": "The approval mode this turn ran in: how much the agent may do without asking you. In Auto-approve the reviewer is named too.",
+ "stamp": "Where and how this turn ran: the machine's name and app version, the sandbox, ⓕ the coworker fingerprint (a 6-character code of the coworker's instructions), then the model and the provider whose key was used.",
+ "turn": "A turn is one message from you and everything the agent did until it finished. Use ‹ › to move between turns.",
+ "tile": {
+ "total": "How long the turn took, from your message to the end of the answer.",
+ "waiting": "Time the agent waited for you: an approval, a question or a plan.",
+ "tools": "Time tools were running, including a helper agent's work inside an explore call.",
+ "model": "Time in model calls for the main work and helpers. Time that overlaps a tool or a wait is counted there, once.",
+ "auto_approver": "Time the auto-approver spent checking actions.",
+ "summary": "Time spent summarising a long conversation so it fits the model.",
+ "sandbox": "Time spent starting the sandbox in this turn.",
+ "other": "The rest of the turn, in none of the tiles before it: the session-title call, building the context, and the app's own work between steps (saving, preparing the next call). Shown when it is 1% of the turn or more.",
+ "offline": "Time inside this turn when this machine had lost its link to the controller. It can overlap other tiles.",
+ "nobody": "Time inside this turn when no window was open on this session. It can overlap other tiles.",
+ "helpers": "How many helper agents (explore) this turn ran.",
+ "helper_tokens": "Tokens the helpers used. They are already inside the session's totals, counted once."
+ },
+ "axis": {
+ "step": "Each row is one step of the turn, in the order they started. Click a row for its details.",
+ "time": "How long the step took. \"<0.1s\" means under a tenth of a second."
+ },
+ "kind": {
+ "context": "OpenWorker put the context together for this turn: instructions, memories, folders, skills and notes.",
+ "model_main": "A call to the model for the main work: it reads the conversation and answers or asks for a tool.",
+ "model_title": "A call that writes the session's name. It runs alongside the main work.",
+ "model_helper": "A call made by a helper agent (explore) that the main agent handed a search to.",
+ "model_summary": "The call that writes a summary of a long conversation.",
+ "model_auto_approver": "The auto-approver's own model call, judging one action.",
+ "model_proxy": "A call that came in through this machine's model address from outside a session.",
+ "model_unattributed": "A model call whose purpose the app could not tell.",
+ "tool": "A tool the agent ran. Click for what it did, who allowed it and what it returned.",
+ "wait": "The agent paused for you: an approval, a question or a plan. \"parked in Inbox\" means nobody was watching the session.",
+ "auto_approver": "The auto-approver checked one action and allowed it, blocked it or asked you.",
+ "shadow": "A shadow check: the reviewer's advice only, recorded next to your own decision. It changes nothing.",
+ "summary": "The conversation was summarised to fit the model; context before → after.",
+ "trim": "Summarising failed, so older messages were cut to make room.",
+ "sandbox": "The sandbox started in this turn; its stages are listed under it.",
+ "helper": "A helper agent's work under its explore call. Click to show or hide its steps."
+ },
+ "notice": {
+ "error": "The turn hit an error.",
+ "interrupted": "You pressed Stop.",
+ "reviewer_paused": "The auto-approver refused three times in a row, so Auto-approve paused.",
+ "step_limit": "The agent used all the steps it may take in one turn.",
+ "truncated": "The model's answer hit its length limit.",
+ "context_changed": "The text OpenWorker adds each turn changed mid-turn (for example a skill became available).",
+ "stamp_changed": "Something changed mid-turn: who is at the app, the auto-approver, or the sandbox. Click for what changed.",
+ "summary_failed": "Writing a summary failed.",
+ "recording_gap": "The app stopped while this turn ran, so part of it was not recorded.",
+ "sandbox_restarted": "The sandbox restarted (or its folder access changed) because the session's folders changed or its runner came back.",
+ "controller_offline": "This machine lost its link to its controller for this long.",
+ "nobody_watching": "No window was open on this session for this long.",
+ "clock_jump": "The computer slept for this long. Steps after it are placed by the wall clock again."
+ },
+ "detail": {
+ "started": "When the step started, counted from the start of the turn (minutes:seconds).",
+ "duration": "How long the step took.",
+ "session_prompt": "A 6-character fingerprint of the coworker's starting instructions (the system prompt). Same text, same code; it stays the same all session. The text itself is not shown.",
+ "attempts": "How many times this request was sent to the provider. More than 1 means it failed and was sent again.",
+ "served": "The exact model version the provider used, shown when it differs from the one picked.",
+ "coworker": "A 6-character fingerprint of which coworker ran: its id, its instructions and its own skill files. The same code means the same coworker version.",
+ "tools": "How many tool definitions were sent to the model. \"set·…\" is a fingerprint of the whole list: the same code on two calls means the same tools were offered.",
+ "request_id": "The provider's own id for this call, from its reply. Quote it to the provider's support to look the call up.",
+ "response_id": "The provider's id for the reply itself.",
+ "error": "Why the call failed: the type and the provider's short message.",
+ "ran_for": "How long the tool itself ran, without any wait for approval.",
+ "waited": "How long the approval card waited for an answer.",
+ "result_size": "The size of what the tool returned to the model: B is bytes, then KB and MB.",
+ "exit": "The command's exit code. 0 means success; anything else means it failed.",
+ "origin": "Who or what allowed (or refused) this call: for example \"low risk\" (no approval needed), \"you\", \"the auto-approver\" or a saved rule.",
+ "blocked": "In a sandbox: how many connections it refused during this call. \"unknown\" when the sandbox cannot say.",
+ "where": "Where the card was shown: in the open session, or in the Inbox because nobody was watching.",
+ "outcome": "How the wait ended: approved, denied, answered, stopped…",
+ "answered_by": "Who answered the card.",
+ "summary_reason": "Why the summary happened: the conversation reached the size set in Settings ▸ Context optimization, or the model said it was too long.",
+ "tokens_before": "The context size in tokens before the summary.",
+ "tokens_after": "The context size in tokens after the summary.",
+ "sbx_provider": "Which sandbox ran the commands.",
+ "sbx_trigger": "When it started: at the start of the turn, or when a tool first needed it.",
+ "sbx_reason": "Why the sandbox restarted.",
+ "sbx_how": "What the restart did to running shells and variables.",
+ "gap_from": "When the gap began, on the wall clock.",
+ "gap_until": "When the gap ended, on the wall clock.",
+ "gap_skipped": "How long the computer slept.",
+ "gap_why": "Why the gap happened.",
+ "result": "What the reviewer decided about this action.",
+ "reason": "The reviewer's own short explanation.",
+ "took": "How long the check took.",
+ "check_tokens": "Tokens used by the reviewer's own model call.",
+ "breaker": "Refusals in a row. At 3, Auto-approve pauses and asks you instead.",
+ "would": "What the reviewer would have done. Advice only: it decided nothing.",
+ "you_decided": "What you decided on the card, and after how long.",
+ "match": "Whether the reviewer's advice matched your decision."
+ },
+ "sec": {
+ "tokens": "Tokens are the units models count text in (about ¾ of a word). \"in\" is everything sent, \"out\" is the reply.",
+ "context_added": "What OpenWorker put into the model's context on this turn besides your message. Counts and codes only, not the text.",
+ "context_now": "The context as it is after the change.",
+ "input": "The messages the model was given, as \"role · text\", each cut to 600 characters. The last 3 are shown; \"Show all\" shows every one.",
+ "output": "What the model answered, and the tools it asked for.",
+ "exact": "The full request and the raw reply, exactly as exchanged with the provider, secrets removed: system text, every tool definition, settings and ids.",
+ "how_allowed": "The approval path, in order: Permission (rules and risk), Grant (something allowed earlier), Auto-approver, Card (asked you) and Answer, each with how it ended.",
+ "arguments": "What the model asked the tool to do.",
+ "result": "What the tool returned to the model (cut if long).",
+ "joined": "OpenWorker keeps a separate audit log of every tool call. These are its entries for this same call: when it was proposed, started and finished.",
+ "inside": "The steps nested inside this one, such as the model call that wrote a summary.",
+ "stages": "The sandbox's start, stage by stage, in the order they ran.",
+ "error_text": "The sandbox's own error message.",
+ "questions": "The questions a decision-model reviewer answered, with yes or no and a score.",
+ "shown": "The text the reviewer was given to judge.",
+ "changed": "What changed mid-turn, with the new values."
+ },
+ "ctx": {
+ "memory": "How many memories the coworker has. \"as an index\" means only a list was given, and the model reads one when it needs it.",
+ "rules": "How many of your standing instructions (Settings ▸ Memory ▸ Your instructions) were included.",
+ "agents_md": "How many AGENTS.md files from the session's folder were included.",
+ "skill": "How many skills the menu offered this turn; ⓢ is a fingerprint of that menu.",
+ "folder": "How many folders the agent was told it can use.",
+ "mode": "A note about the current mode (Discuss or Plan) was included.",
+ "sandbox": "A paragraph about the sandbox was included.",
+ "block": "The text OpenWorker attaches to your message each turn (the <system-context> in the Input box): its length in characters and its fingerprint. If it changes mid-turn, a \"Context changed\" row appears."
+ },
+ "tok": {
+ "cache": "Input the provider served from its cache because it saw this text recently. Cheaper and faster.",
+ "written": "Input the provider stored in its cache for next time. The provider reports this number; \"—\" means it did not say.",
+ "new": "Input neither read from nor written to the cache: the provider's input total minus the other two.",
+ "out": "Output tokens of the reply. \"thinking\" is the part spent on reasoning."
+ },
+ "usage": {
+ "input": "All input tokens of the session: fresh, read from cache and written to cache. \"≥\" means at least, because some calls did not report.",
+ "output": "All output tokens of the session, reasoning included.",
+ "cache": "The share of input the provider served from its cache.",
+ "purpose": "Tokens by what the calls were for: the main work, helper agents, summaries, the auto-approver and the session title.",
+ "model": "Tokens by model. In: all input. Cache: input read from cache. Out: output.",
+ "context": "How full the model's context window is after the last main call."
+ },
+ "issues": {
+ "information": "Advice, not a warning: it does not count on the tab or light the sidebar ring.",
+ "fine": "Checks that ran on this turn and found nothing.",
+ "no_plan": "The \"Left its plan\" check needs a to-do list, and the agent kept none in this turn.",
+ "summary": "Every turn is checked against 19 rules. Some only apply in certain situations (a to-do list, a sandbox, Auto-approve), so the number that ran changes from turn to turn.",
+ "na": "Checks that did not apply to this turn, and why."
+ },
+ "issue": {
+ "repeated_step": "The same tool with the same arguments ran 3 or more times in a row.",
+ "off_plan": "The to-do list lost an unfinished item, or the agent kept acting after marking every item done.",
+ "refused": "You, a rule or the auto-approver said no to an action.",
+ "overridden": "You used \"Allow anyway\" on something that was blocked.",
+ "reviewer_could_not_run": "The auto-approver failed, so you were asked instead.",
+ "reviewer_paused": "Three refusals in a row paused Auto-approve.",
+ "reviewer_allowed_send": "The auto-approver allowed an action that reaches the network.",
+ "shadow_disagreed": "The advice-only reviewer would have decided differently from you.",
+ "tool_failed": "A tool errored, returned an error, or a command exited with a non-zero code.",
+ "model_failed": "A main model call failed and was not recovered in the turn.",
+ "retried": "A model call needed more than one try.",
+ "cut_off": "The model's answer hit its length limit.",
+ "step_limit": "The agent used all the steps it may take in one turn.",
+ "made_up_tool": "The model asked for a tool that does not exist, or gave arguments that do not fit the tool.",
+ "context_full": "The conversation outgrew the model's window: summarised in a hurry, older messages cut, a summary failed, or 95% full.",
+ "blocked_by_sandbox": "The sandbox refused a connection this command tried.",
+ "sandbox_failed": "The sandbox could not start.",
+ "ran_without_sandbox": "This machine expects a sandbox, but this turn ran actions without one.",
+ "waited_unwatched": "An approval, question or plan waited 15 minutes or more while nobody was at the app."
+ }
+ },
"gap": {
"offline_total": "Machine offline",
"nobody_total": "Nobody watching",
@@ -3114,6 +3280,79 @@
"sandbox_failed": "The sandbox could not start",
"ran_without_sandbox": "Ran without the sandbox",
"waited_unwatched": "Waited while nobody was watching"
+ },
+ "summary": "{{total}} checks: {{fired}} found something · {{fine}} fine · {{na}} didn't apply",
+ "na": "Didn't apply · {{n}}",
+ "na_reason": {
+ "off_plan": "no to-do list in this turn",
+ "overridden": "only in Auto-approve mode",
+ "reviewer_could_not_run": "only in Auto-approve mode or with Shadow evaluation on",
+ "reviewer_paused": "only in Auto-approve mode",
+ "reviewer_allowed_send": "only in Auto-approve mode",
+ "shadow_disagreed": "only with Shadow evaluation on",
+ "blocked_by_sandbox": "only in a sandbox that enforces its limits",
+ "sandbox_failed": "only in a turn that started a sandbox",
+ "ran_without_sandbox": "only on a machine that expects a sandbox"
+ },
+ "say": {
+ "repeated_step": "{{name}} was called {{n}} times in a row with the same arguments. That cost {{extra}} extra tokens.",
+ "off_plan_dropped": "A plan step was dropped before it was done ({{dropped}} of {{of}}).",
+ "off_plan_after_done": "It kept acting after it marked every plan step done: {{name}}.",
+ "refused_user": "You refused {{name}}.",
+ "refused_denied_by_rule": "A rule refused {{name}}.",
+ "refused_reviewer_denied": "The auto-approver refused {{name}} and you did not allow it.",
+ "refused_unattended_auto": "{{name}} was refused because nobody was at the app to approve it.",
+ "overridden": "The auto-approver blocked {{name}} and you used Allow anyway.",
+ "reviewer_could_not_run": "The auto-approver could not check {{name}}, so it was shown to you instead.",
+ "reviewer_paused": "The auto-approver paused after {{denials}} refusals in a row. Later actions were shown to you.",
+ "reviewer_allowed_send": "The auto-approver allowed {{name}} without asking you. It sends data off the machine.",
+ "shadow_disagreed": "The shadow reviewer would have said \"{{result}}\" on {{name}}.",
+ "tool_failed": "{{name}} failed.",
+ "tool_failed_exit": "{{name}} failed with exit code {{exit}}.",
+ "tool_failed_recovered": "A later {{name}} in this turn worked.",
+ "model_failed": "The model call failed ({{error}}).",
+ "model_failed_stopped": "The turn stopped here.",
+ "retried": "The model call needed {{attempts}} tries ({{purpose}}).",
+ "cut_off": "The model stopped at its output limit. The answer may be incomplete.",
+ "step_limit": "The turn stopped after {{iterations}} steps, the limit.",
+ "made_up_tool_unknown_tool": "The model asked for a tool that does not exist: {{name}}.",
+ "made_up_tool_bad_arguments": "The model sent arguments that do not fit {{name}}.",
+ "context_full_trim": "Older messages were cut to make room.",
+ "context_full_overflow": "The conversation became too long for the model, so it was summarised in a hurry.",
+ "context_full_summary_failed": "Summarising failed, so the context stayed full.",
+ "context_full_hard_fill": "The context was {{pct}}% full on this call.",
+ "blocked_by_sandbox": "The sandbox blocked {{n}} network request(s) during {{name}}.",
+ "blocked_by_sandbox_exit": "The sandbox blocked {{n}} network request(s) during {{name}}, which exited with code {{exit}}.",
+ "sandbox_failed": "The sandbox could not start at {{stage}}: {{type}}.",
+ "ran_without_sandbox": "This machine has a sandbox switched on, but this turn ran {{n}} action(s) without it.",
+ "waited_unwatched": "It waited {{minutes}} min for an answer while nobody was at the app.",
+ "waited_unwatched_inbox": "It waited {{minutes}} min in the Inbox for an answer while nobody was at the app."
+ },
+ "fine_say": {
+ "repeated_step": "No step repeated {{min_run}} times in a row",
+ "off_plan": "No plan step was dropped ({{done}} of {{items}} done)",
+ "refused": "No actions were refused by you, a rule or the auto-approver",
+ "overridden": "You did not override the auto-approver",
+ "reviewer_paused": "Auto-approver was never paused",
+ "reviewer_allowed_send": "The auto-approver let nothing leave the machine without asking you",
+ "shadow_disagreed": "The shadow reviewer agreed with you on {{n}} of {{n}} actions",
+ "tool_failed": "No tool failed",
+ "model_failed": "No model call failed",
+ "retried": "No model call needed a second try",
+ "cut_off": "No answers were cut off (length limit)",
+ "step_limit": "Stayed under the step limit ({{iterations}} of {{cap}})",
+ "step_limit_nocap": "Stayed under the step limit ({{iterations}} steps)",
+ "made_up_tool": "No unknown tools or invalid arguments",
+ "context_full": "Context had room (at most {{pct}}% of the window)",
+ "context_full_normal": "; summarised {{n}} time(s) at the normal point",
+ "blocked_by_sandbox": "The sandbox blocked nothing",
+ "sandbox_failed": "The sandbox started ({{provider}}, {{seconds}} s)",
+ "ran_without_sandbox": "Ran inside the sandbox",
+ "waited_unwatched": "Nothing waited long for an answer while nobody was watching",
+ "reviewer_could_not_run_live_one": "Auto-approver checked {{count}} action: {{allowed}} allowed, {{asked}} asked you, {{blocked}} blocked; none failed to run",
+ "reviewer_could_not_run_live_other": "Auto-approver checked {{count}} actions: {{allowed}} allowed, {{asked}} asked you, {{blocked}} blocked; none failed to run",
+ "reviewer_could_not_run_shadow_one": "Shadow reviewer checked {{count}} action; none failed to run",
+ "reviewer_could_not_run_shadow_other": "Shadow reviewer checked {{count}} actions; none failed to run"
}
},
"title": "Steps",
@@ -3170,10 +3409,9 @@
"how_allowed": "How it was allowed",
"arguments": "Arguments",
"result": "Result",
- "joined": "Joined records",
- "joined_line_one": "{{count}} audit row",
- "joined_line_other": "{{count}} audit rows",
- "same_call": "same call id",
+ "joined": "Audit log",
+ "joined_line_one": "{{count}} entry for this call: {{stages}}",
+ "joined_line_other": "{{count}} entries for this call: {{stages}}",
"inside": "Inside this step",
"check_title": "Auto-approver check",
"shadow_title": "Shadow check",
@@ -3226,7 +3464,9 @@
"auto_approver": "Auto-approver",
"summary": "Summary",
"other": "Other",
- "sandbox": "Sandbox setup"
+ "sandbox": "Sandbox setup",
+ "helpers": "Helpers",
+ "helper_tokens": "Helper tokens"
},
"checks_line_one": "Auto-approve: {{count}} check · {{allowed}} allowed · {{asked}} asked you · {{blocked}} blocked · {{time}} · {{reviewer}}",
"checks_line_other": "Auto-approve: {{count}} checks · {{allowed}} allowed · {{asked}} asked you · {{blocked}} blocked · {{time}} · {{reviewer}}",
@@ -3253,7 +3493,8 @@
"none": "No sandbox",
"full": "enforced",
"partial": "partly enforced",
- "network": "network {{profile}}"
+ "network": "network {{profile}}",
+ "not_enforced": "not enforced"
},
"row": {
"context": "Context built",
@@ -3634,7 +3875,9 @@
"last_sent": "Last sent {{when}}",
"counts": "Today: {{sent}} steps sent · {{failed}} failed · {{dropped}} dropped.",
"resumes": "If the destination stops answering, nothing is lost: sending resumes where it stopped.",
- "failed_sub": "Sending has stopped until the settings change."
+ "failed_sub": "Sending has stopped until the settings change.",
+ "outside_one": "Today: {{count}} model call from outside any session ({{tokens}} tokens), through this machine's model address. They are not in any session's Steps or Usage.",
+ "outside_other": "Today: {{count}} model calls from outside any session ({{tokens}} tokens), through this machine's model address. They are not in any session's Steps or Usage."
},
"state": {
"off": "Off",
diff --git a/surfaces/gui/src/locales/zh.json b/surfaces/gui/src/locales/zh.json
index 1ff7bcf1..1d0ab7b1 100644
--- a/surfaces/gui/src/locales/zh.json
+++ b/surfaces/gui/src/locales/zh.json
@@ -2993,6 +2993,172 @@
}
},
"steps": {
+ "help": {
+ "mode": "本回合使用的审批模式:智能体无需询问你即可做多少事。自动审批模式下还会显示审阅者。",
+ "stamp": "本回合在哪里、如何运行:机器名称和应用版本、沙箱、ⓕ 协作者指纹(协作者指令的 6 位代码),以及所用模型和提供方密钥。",
+ "turn": "一个回合是你的一条消息,以及智能体直到完成为止所做的一切。用 ‹ › 在回合间切换。",
+ "tile": {
+ "total": "本回合用时:从你的消息到回答结束。",
+ "waiting": "智能体等待你的时间:审批、问题或计划。",
+ "tools": "工具运行的时间,包括 explore 调用中辅助智能体的工作。",
+ "model": "主要工作和辅助智能体的模型调用时间。与工具或等待重叠的时间只计入那里一次。",
+ "auto_approver": "自动审批检查操作所用的时间。",
+ "summary": "为让长对话适配模型而进行摘要的时间。",
+ "sandbox": "本回合启动沙箱所用的时间。",
+ "other": "本回合中不属于前面任何一项的时间:会话标题调用、构建上下文,以及应用在步骤之间的工作(保存、准备下一次调用)。占本回合 1% 或以上时显示。",
+ "offline": "本回合中这台机器与控制端断开连接的时间。可能与其他项重叠。",
+ "nobody": "本回合中没有窗口打开此会话的时间。可能与其他项重叠。",
+ "helpers": "本回合运行了多少个辅助智能体(explore)。",
+ "helper_tokens": "辅助智能体使用的 token。已包含在会话总数中,只计一次。"
+ },
+ "axis": {
+ "step": "每一行是本回合的一个步骤,按开始顺序排列。点击一行查看详情。",
+ "time": "该步骤用时。\"<0.1s\" 表示不到十分之一秒。"
+ },
+ "kind": {
+ "context": "OpenWorker 为本回合组装上下文:指令、记忆、文件夹、技能和说明。",
+ "model_main": "用于主要工作的模型调用:读取对话后回答或请求使用工具。",
+ "model_title": "为会话命名的调用,与主要工作同时进行。",
+ "model_helper": "辅助智能体(explore)发出的调用,主智能体把搜索交给了它。",
+ "model_summary": "为长对话撰写摘要的调用。",
+ "model_auto_approver": "自动审批自身的模型调用,用于评判一个操作。",
+ "model_proxy": "从会话之外通过本机模型地址发来的调用。",
+ "model_unattributed": "应用无法判断用途的模型调用。",
+ "tool": "智能体运行的工具。点击查看它做了什么、由谁允许以及返回了什么。",
+ "wait": "智能体在等你:审批、问题或计划。\"已放入收件箱\" 表示当时没人查看此会话。",
+ "auto_approver": "自动审批检查了一个操作:允许、阻止或询问你。",
+ "shadow": "影子检查:仅记录审阅者的建议,与你的决定并列,不改变任何事。",
+ "summary": "对话被摘要以适配模型;上下文 前 → 后。",
+ "trim": "摘要失败,因此删去了较早的消息以腾出空间。",
+ "sandbox": "沙箱在本回合启动;其阶段列在下方。",
+ "helper": "explore 调用下辅助智能体的工作。点击展开或收起其步骤。"
+ },
+ "notice": {
+ "error": "本回合出现错误。",
+ "interrupted": "你按下了停止。",
+ "reviewer_paused": "自动审批连续拒绝三次,因此已暂停。",
+ "step_limit": "智能体用完了一个回合可走的全部步骤。",
+ "truncated": "模型的回答达到了长度上限。",
+ "context_changed": "OpenWorker 每回合附加的文字在回合中发生了变化(例如有新技能可用)。",
+ "stamp_changed": "回合中有设置变化:是否有人在应用前、自动审批或沙箱。点击查看变化内容。",
+ "summary_failed": "撰写摘要失败。",
+ "recording_gap": "本回合运行时应用停止了,因此部分内容没有记录。",
+ "sandbox_restarted": "由于会话文件夹变化或运行器重新连接,沙箱已重启(或文件夹访问权限已更改)。",
+ "controller_offline": "这台机器与控制端断开连接的时长。",
+ "nobody_watching": "没有窗口打开此会话的时长。",
+ "clock_jump": "电脑休眠的时长。之后的步骤重新按系统时钟排列。"
+ },
+ "detail": {
+ "started": "步骤开始时间,从本回合开始算起(分:秒)。",
+ "duration": "该步骤用时。",
+ "session_prompt": "协作者初始指令(系统提示)的 6 位指纹。文字相同,代码就相同;整个会话保持不变。不显示文字本身。",
+ "attempts": "此请求发送给提供方的次数。大于 1 表示失败后重发。",
+ "served": "提供方实际使用的模型版本,与所选不同时显示。",
+ "coworker": "运行的协作者的 6 位指纹:其标识、指令和自带技能文件。代码相同表示协作者版本相同。",
+ "tools": "发送给模型的工具定义数量。\"set·…\" 是整份列表的指纹:两次调用代码相同,说明提供的工具相同。",
+ "request_id": "提供方为此调用给出的编号,来自其回复。联系提供方支持时可用它查找此调用。",
+ "response_id": "提供方为回复本身给出的编号。",
+ "error": "调用失败的原因:类型和提供方的简短说明。",
+ "ran_for": "工具本身运行的时长,不含等待审批。",
+ "waited": "审批卡片等待回答的时长。",
+ "result_size": "工具返回给模型的内容大小:B 为字节,然后是 KB 和 MB。",
+ "exit": "命令的退出码。0 表示成功,其他值表示失败。",
+ "origin": "由谁或什么允许(或拒绝)了此调用:例如 \"低风险\"(无需审批)、\"你\"、\"自动审批\" 或已保存的规则。",
+ "blocked": "在沙箱中:此调用期间被拒绝的连接数。沙箱无法说明时显示 \"未知\"。",
+ "where": "卡片显示的位置:在打开的会话中,或因无人查看而在收件箱中。",
+ "outcome": "等待如何结束:已批准、已拒绝、已回答、已停止……",
+ "answered_by": "回答卡片的人。",
+ "summary_reason": "摘要的原因:对话达到了 设置 ▸ 上下文优化 中设定的大小,或模型表示对话过长。",
+ "tokens_before": "摘要前的上下文大小(token)。",
+ "tokens_after": "摘要后的上下文大小(token)。",
+ "sbx_provider": "运行命令的沙箱。",
+ "sbx_trigger": "启动时机:回合开始时,或某个工具首次需要时。",
+ "sbx_reason": "沙箱重启的原因。",
+ "sbx_how": "重启对正在运行的终端和变量的影响。",
+ "gap_from": "缺口开始的时间(系统时钟)。",
+ "gap_until": "缺口结束的时间(系统时钟)。",
+ "gap_skipped": "电脑休眠的时长。",
+ "gap_why": "出现缺口的原因。",
+ "result": "审阅者对此操作的决定。",
+ "reason": "审阅者自己的简短说明。",
+ "took": "检查所用时间。",
+ "check_tokens": "审阅者自身模型调用所用的 token。",
+ "breaker": "连续拒绝次数。达到 3 次时,自动审批暂停并改为询问你。",
+ "would": "审阅者本来会怎么做。仅为建议,不做任何决定。",
+ "you_decided": "你在卡片上的决定,以及用了多长时间。",
+ "match": "审阅者的建议是否与你的决定一致。"
+ },
+ "sec": {
+ "tokens": "token 是模型计算文字的单位(约为一个单词的四分之三)。\"in\" 是发送的全部内容,\"out\" 是回复。",
+ "context_added": "本回合除你的消息外,OpenWorker 放入模型上下文的内容。只有数量和代码,不含文字。",
+ "context_now": "变化之后的上下文。",
+ "input": "模型收到的消息,格式为 \"角色 · 文字\",每条截取 600 个字符。默认显示最后 3 条;\"显示全部\" 显示所有。",
+ "output": "模型的回答,以及它请求使用的工具。",
+ "exact": "与提供方交换的完整请求和原始回复(已去除密钥):系统文字、每个工具定义、设置和编号。",
+ "how_allowed": "审批路径,按顺序:权限(规则和风险)、授权(之前已允许的)、自动审批、卡片(询问你)和回答,各自显示结果。",
+ "arguments": "模型要求工具做的事。",
+ "result": "工具返回给模型的内容(过长时截断)。",
+ "joined": "OpenWorker 为每次工具调用另外保留一份审计日志。这些是同一调用的记录:何时被提议、开始和完成。",
+ "inside": "嵌套在此步骤内的步骤,例如撰写摘要的模型调用。",
+ "stages": "沙箱的启动过程,按运行顺序逐个阶段列出。",
+ "error_text": "沙箱自己的错误信息。",
+ "questions": "决策模型审阅者回答的问题,以及是/否和分数。",
+ "shown": "审阅者用于评判的文字。",
+ "changed": "回合中发生的变化及新值。"
+ },
+ "ctx": {
+ "memory": "协作者拥有的记忆数量。\"作为索引\" 表示只提供了列表,模型需要时再读取。",
+ "rules": "包含了多少条你的常设指令(设置 ▸ 记忆 ▸ 你的指令)。",
+ "agents_md": "包含了会话文件夹中多少个 AGENTS.md 文件。",
+ "skill": "本回合技能菜单提供的技能数量;ⓢ 是该菜单的指纹。",
+ "folder": "告知智能体可以使用的文件夹数量。",
+ "mode": "包含了关于当前模式(讨论或计划)的说明。",
+ "sandbox": "包含了一段关于沙箱的说明。",
+ "block": "OpenWorker 每回合附加到你消息上的文字(输入框中的 <system-context>):其字符长度和指纹。若在回合中变化,会出现 \"上下文已变化\" 一行。"
+ },
+ "tok": {
+ "cache": "提供方因最近见过这段文字而从缓存提供的输入。更便宜、更快。",
+ "written": "提供方存入缓存以备下次使用的输入。此数字由提供方报告;\"—\" 表示未报告。",
+ "new": "既未从缓存读取也未写入缓存的输入:提供方的输入总数减去前两项。",
+ "out": "回复的输出 token。\"思考\" 是用于推理的部分。"
+ },
+ "usage": {
+ "input": "本会话的全部输入 token:新输入、缓存读取和缓存写入。\"≥\" 表示至少,因为有些调用未报告。",
+ "output": "本会话的全部输出 token,包括推理。",
+ "cache": "提供方从缓存提供的输入所占比例。",
+ "purpose": "按调用用途统计的 token:主要工作、辅助智能体、摘要、自动审批和会话标题。",
+ "model": "按模型统计的 token。输入:全部输入。缓存:从缓存读取的输入。输出:输出。",
+ "context": "最后一次主要调用后,模型上下文窗口的占用程度。"
+ },
+ "issues": {
+ "information": "建议而非警告:不计入标签数量,也不会点亮侧栏圆环。",
+ "fine": "本回合运行过但没有发现问题的检查。",
+ "no_plan": "\"偏离计划\" 检查需要待办列表,而智能体本回合没有建立。",
+ "summary": "每个回合都会按 19 条规则检查。有些只在特定情况下适用(待办列表、沙箱、自动审批),所以每个回合运行的数量会不同。",
+ "na": "不适用于本回合的检查,以及原因。"
+ },
+ "issue": {
+ "repeated_step": "同一个工具以相同参数连续运行了 3 次或更多。",
+ "off_plan": "待办列表丢失了未完成的条目,或在标记全部完成后智能体仍继续操作。",
+ "refused": "你、规则或自动审批拒绝了某个操作。",
+ "overridden": "你对已被阻止的操作使用了 \"仍然允许\"。",
+ "reviewer_could_not_run": "自动审批失败,因此改为询问你。",
+ "reviewer_paused": "连续三次拒绝使自动审批暂停。",
+ "reviewer_allowed_send": "自动审批允许了一个会访问网络的操作。",
+ "shadow_disagreed": "仅提供建议的审阅者本会做出与你不同的决定。",
+ "tool_failed": "工具出错、返回了错误,或命令以非零退出码结束。",
+ "model_failed": "一次主要模型调用失败,且本回合内未恢复。",
+ "retried": "一次模型调用尝试了不止一次。",
+ "cut_off": "模型的回答达到了长度上限。",
+ "step_limit": "智能体用完了一个回合可走的全部步骤。",
+ "made_up_tool": "模型请求了不存在的工具,或给出了与工具不符的参数。",
+ "context_full": "对话超出了模型窗口:被仓促摘要、删去较早消息、摘要失败,或占用达 95%。",
+ "blocked_by_sandbox": "沙箱拒绝了此命令尝试的连接。",
+ "sandbox_failed": "沙箱无法启动。",
+ "ran_without_sandbox": "这台机器应使用沙箱,但本回合在没有沙箱的情况下执行了操作。",
+ "waited_unwatched": "审批、问题或计划在无人查看时等待了 15 分钟或更久。"
+ }
+ },
"gap": {
"offline_total": "机器离线",
"nobody_total": "无人查看",
@@ -3069,6 +3235,77 @@
"sandbox_failed": "沙盒无法启动",
"ran_without_sandbox": "未在沙盒中运行",
"waited_unwatched": "无人值守时等待"
+ },
+ "summary": "共 {{total}} 项检查:{{fired}} 项发现问题 · {{fine}} 项正常 · {{na}} 项不适用",
+ "na": "不适用 · {{n}}",
+ "na_reason": {
+ "off_plan": "本回合没有待办列表",
+ "overridden": "仅在自动审批模式下",
+ "reviewer_could_not_run": "仅在自动审批模式下或开启影子评估时",
+ "reviewer_paused": "仅在自动审批模式下",
+ "reviewer_allowed_send": "仅在自动审批模式下",
+ "shadow_disagreed": "仅在开启影子评估时",
+ "blocked_by_sandbox": "仅在强制执行限制的沙箱中",
+ "sandbox_failed": "仅在启动了沙箱的回合中",
+ "ran_without_sandbox": "仅在应使用沙箱的机器上"
+ },
+ "say": {
+ "repeated_step": "{{name}} 以相同参数连续被调用了 {{n}} 次,多用了 {{extra}} 个 token。",
+ "off_plan_dropped": "有计划步骤在完成前被删除({{of}} 项中的 {{dropped}} 项)。",
+ "off_plan_after_done": "在把所有计划步骤标为完成后,它仍继续操作:{{name}}。",
+ "refused_user": "你拒绝了 {{name}}。",
+ "refused_denied_by_rule": "规则拒绝了 {{name}}。",
+ "refused_reviewer_denied": "自动审批拒绝了 {{name}},你也没有允许。",
+ "refused_unattended_auto": "{{name}} 被拒绝,因为没有人在应用前批准。",
+ "overridden": "自动审批阻止了 {{name}},而你使用了 \"仍然允许\"。",
+ "reviewer_could_not_run": "自动审批无法检查 {{name}},因此改为询问你。",
+ "reviewer_paused": "自动审批在连续 {{denials}} 次拒绝后暂停。之后的操作都询问了你。",
+ "reviewer_allowed_send": "自动审批未询问你就允许了 {{name}}。它会把数据发送到本机之外。",
+ "shadow_disagreed": "影子审阅者本会对 {{name}} 选择 \"{{result}}\"。",
+ "tool_failed": "{{name}} 失败。",
+ "tool_failed_exit": "{{name}} 失败,退出码 {{exit}}。",
+ "tool_failed_recovered": "本回合中之后的 {{name}} 成功了。",
+ "model_failed": "模型调用失败({{error}})。",
+ "model_failed_stopped": "本回合在此停止。",
+ "retried": "模型调用尝试了 {{attempts}} 次({{purpose}})。",
+ "cut_off": "模型达到了输出上限,回答可能不完整。",
+ "step_limit": "本回合在 {{iterations}} 步后停止,已达上限。",
+ "made_up_tool_unknown_tool": "模型请求了不存在的工具:{{name}}。",
+ "made_up_tool_bad_arguments": "模型给 {{name}} 的参数与它不符。",
+ "context_full_trim": "删去了较早的消息以腾出空间。",
+ "context_full_overflow": "对话对模型来说太长,因此被仓促摘要。",
+ "context_full_summary_failed": "摘要失败,上下文仍然是满的。",
+ "context_full_hard_fill": "这次调用时上下文已满 {{pct}}%。",
+ "blocked_by_sandbox": "沙箱在 {{name}} 期间阻止了 {{n}} 个网络请求。",
+ "blocked_by_sandbox_exit": "沙箱在 {{name}} 期间阻止了 {{n}} 个网络请求,该命令以退出码 {{exit}} 结束。",
+ "sandbox_failed": "沙箱无法在{{stage}}阶段启动:{{type}}。",
+ "ran_without_sandbox": "这台机器已开启沙箱,但本回合有 {{n}} 个操作在沙箱外运行。",
+ "waited_unwatched": "它在无人查看时等待回答 {{minutes}} 分钟。",
+ "waited_unwatched_inbox": "它在收件箱中、无人查看时等待回答 {{minutes}} 分钟。"
+ },
+ "fine_say": {
+ "repeated_step": "没有步骤连续重复 {{min_run}} 次",
+ "off_plan": "没有删除计划步骤(已完成 {{done}}/{{items}})",
+ "refused": "没有操作被你、规则或自动审批拒绝",
+ "overridden": "你没有推翻自动审批",
+ "reviewer_paused": "自动审批从未暂停",
+ "reviewer_allowed_send": "自动审批没有在未询问你的情况下让任何数据离开本机",
+ "shadow_disagreed": "影子审阅者在 {{n}} 个操作中与你一致 {{n}} 次",
+ "tool_failed": "没有工具失败",
+ "model_failed": "没有模型调用失败",
+ "retried": "没有模型调用需要重试",
+ "cut_off": "没有回答被截断(长度上限)",
+ "step_limit": "未超出步骤上限({{iterations}}/{{cap}})",
+ "step_limit_nocap": "未超出步骤上限({{iterations}} 步)",
+ "made_up_tool": "没有未知工具或无效参数",
+ "context_full": "上下文还有空间(最多占窗口的 {{pct}}%)",
+ "context_full_normal": ";在正常时机摘要了 {{n}} 次",
+ "blocked_by_sandbox": "沙箱没有阻止任何请求",
+ "sandbox_failed": "沙箱已启动({{provider}},{{seconds}} 秒)",
+ "ran_without_sandbox": "在沙箱内运行",
+ "waited_unwatched": "没有在无人查看时长时间等待回答",
+ "reviewer_could_not_run_live_other": "自动审批检查了 {{count}} 个操作:允许 {{allowed}},询问你 {{asked}},阻止 {{blocked}};全部正常运行",
+ "reviewer_could_not_run_shadow_other": "影子审阅者检查了 {{count}} 个操作;全部正常运行"
}
},
"title": "步骤",
@@ -3124,9 +3361,8 @@
"how_allowed": "如何获得允许",
"arguments": "参数",
"result": "结果",
- "joined": "关联记录",
- "joined_line_other": "{{count}} 条审计记录",
- "same_call": "相同调用 id",
+ "joined": "审计日志",
+ "joined_line_other": "此调用有 {{count}} 条记录:{{stages}}",
"inside": "此步骤内部",
"check_title": "自动批准检查",
"shadow_title": "影子检查",
@@ -3161,7 +3397,8 @@
"reviewer_off": "关闭",
"changed_sandbox": "沙盒",
"sites_other": "{{count}} 个站点",
- "runner": "运行器 {{version}}"
+ "runner": "运行器 {{version}}",
+ "joined_line_one": "此调用有 {{count}} 条记录:{{stages}}"
},
"loading": "加载中…",
"load_error": "无法加载步骤。",
@@ -3178,7 +3415,9 @@
"auto_approver": "自动审批",
"summary": "摘要",
"other": "其他",
- "sandbox": "沙箱启动"
+ "sandbox": "沙箱启动",
+ "helpers": "辅助智能体",
+ "helper_tokens": "辅助 token"
},
"checks_line_other": "自动审批:{{count}} 次检查 · {{allowed}} 次允许 · {{asked}} 次询问你 · {{blocked}} 次阻止 · {{time}} · {{reviewer}}",
"unknown_line_one": "{{count}} 个步骤在 OpenWorker 停止时仍在运行,其用时未知。",
@@ -3204,7 +3443,8 @@
"none": "无沙箱",
"full": "强制执行",
"partial": "部分执行",
- "network": "网络 {{profile}}"
+ "network": "网络 {{profile}}",
+ "not_enforced": "未强制"
},
"row": {
"context": "构建上下文",
@@ -3578,7 +3818,9 @@
"last_sent": "上次发送:{{when}}",
"counts": "今天:已发送 {{sent}} 个步骤 · {{failed}} 次失败 · {{dropped}} 个已丢弃。",
"resumes": "如果目标停止响应,不会丢失任何内容:发送会从中断处继续。",
- "failed_sub": "在设置更改之前,发送已停止。"
+ "failed_sub": "在设置更改之前,发送已停止。",
+ "outside_one": "今天:{{count}} 次来自会话之外的模型调用({{tokens}} token),经由本机的模型地址。它们不在任何会话的步骤或用量中。",
+ "outside_other": "今天:{{count}} 次来自会话之外的模型调用({{tokens}} token),经由本机的模型地址。它们不在任何会话的步骤或用量中。"
},
"state": {
"off": "关闭",
diff --git a/surfaces/gui/src/styles.css b/surfaces/gui/src/styles.css
index 4faf19f3..645d9281 100644
--- a/surfaces/gui/src/styles.css
+++ b/surfaces/gui/src/styles.css
@@ -2487,6 +2487,13 @@ html[data-theme="dark"] { --bar-model:#4c8dff; --bar-tool:#3fb8ad; --bar-wait:#3
.k-ctx .steps-kind, .steps-kind.k-ctx, .k-ctx .steps-bar { background:var(--bar-ctx); }
.k-sbx .steps-kind, .steps-kind.k-sbx, .k-sbx .steps-bar { background:var(--bar-sbx); }
.steps-stage { cursor:default; }
+/* Hover and focus help on labels (observability F-6). */
+.help { text-decoration: underline dotted; text-decoration-color: var(--faint); text-underline-offset: 3px; cursor: help; border-radius: 3px; }
+.help:focus-visible { outline: 2px solid var(--accent); outline-offset: 2px; }
+.help-tip { position: fixed; z-index: 1000; width: max-content; max-width: 280px; background: var(--ink); color: var(--canvas);
+ border-radius: 10px; padding: 8px 10px; font-size: var(--fs-meta); font-weight: 400; line-height: 1.45; text-align: left;
+ white-space: normal; text-transform: none; letter-spacing: normal; box-shadow: 0 8px 24px rgba(0, 0, 0, 0.18); pointer-events: none; }
+.help-tip.up { transform: translateY(-100%); }
.k-gap .steps-kind, .steps-kind.k-gap { background:transparent; outline:1px dashed var(--faint); }
.k-gap .steps-bar { background:repeating-linear-gradient(90deg, var(--line-strong) 0 2px, transparent 2px 5px); border:1px dashed var(--faint); top:1px; height:10px; }
.steps-stage .steps-bar { opacity:.75; }