Skip to content

🙌 OpenHands: AI-Driven Development

feat(verify-openhands): hold keys with `browser key down|up|release-all` (#18292)

Engel Nystfddb0ddcfe0d
diff --git a/.agents/skills/verify-openhands/references/mapping.md b/.agents/skills/verify-openhands/references/mapping.md
index c8799239ad..3cffaa7681 100644
--- a/.agents/skills/verify-openhands/references/mapping.md
+++ b/.agents/skills/verify-openhands/references/mapping.md
@@ -40,7 +40,10 @@ candidate. Sources, roughly in order of yield:
controls, and anything the agent can drive through client tools
(`canvas_ui_control`).
5. **Keyboard shortcuts**: grep `keydown`, `useHotkeys`, `metaKey`, `ctrlKey` in
- `src/`. A shortcut is an entry point.
+ `src/`. A shortcut is an entry point. When it reads `event.repeat` (a held
+ key must not fire twice) or needs a modifier held across a click, drive it
+ with `browser key down <Key> [--repeat N]` and `browser key up <Key>`
+ (`browser press` sends one keydown and keyup).
6. **Backends and modes**: local vs Cloud backend, public vs local auth, ACP
agents, partial stacks (`agent-canvas --help`), Electron, Docker, and the
embeddable library (`src/index.ts` / `build:lib`).
diff --git a/.agents/skills/verify-openhands/scripts/browser-daemon.mjs b/.agents/skills/verify-openhands/scripts/browser-daemon.mjs
index f3acdbb62c..52d01ec779 100755
--- a/.agents/skills/verify-openhands/scripts/browser-daemon.mjs
+++ b/.agents/skills/verify-openhands/scripts/browser-daemon.mjs
@@ -22,6 +22,7 @@ import { createRequire } from "node:module";
import { basename, dirname, extname, join, resolve } from "node:path";
import { fileURLToPath } from "node:url";
import { redactStorage } from "./lib/redact-storage.mjs";
+import { HeldKeys } from "./lib/held-keys.mjs";
import { SECRET_KEY, SECRET_PATH, redactBody } from "./lib/network-bodies.mjs";
import { buildLocator, toCss } from "./lib/selectors.mjs";
import {
@@ -190,6 +191,8 @@ function redactQuery(search) {
}
let markIndex = 0;
let activePage = context.pages()[0] ?? (await context.newPage());
+// Keys held with `browser key down` until `key up` or `key release-all`.
+const heldKeys = new HeldKeys();
function originOf(url) {
try {
@@ -766,6 +769,15 @@ const handlers = {
else await activePage.keyboard.press(key);
return { url: activePage.url() };
},
+ async "key-down"({ key, repeat, intervalMs }) {
+ return heldKeys.down(activePage, key, { repeat, intervalMs });
+ },
+ async "key-up"({ key }) {
+ return heldKeys.up(key);
+ },
+ async "key-release-all"() {
+ return heldKeys.releaseAll();
+ },
async check({ selector, timeout }) {
await locate(selector).check({ timeout });
return { checked: true };
diff --git a/.agents/skills/verify-openhands/scripts/control-openhands.mjs b/.agents/skills/verify-openhands/scripts/control-openhands.mjs
index 96d45ba8da..cd0af5710f 100755
--- a/.agents/skills/verify-openhands/scripts/control-openhands.mjs
+++ b/.agents/skills/verify-openhands/scripts/control-openhands.mjs
@@ -42,6 +42,7 @@ import {
} from "./lib/testids.mjs";
import { tmuxPathFor } from "./lib/tmux-path.mjs";
import { browserCallLimit } from "./lib/call-limit.mjs";
+import { keyDownArguments } from "./lib/held-keys.mjs";
import {
DEFAULT_AGENT_PROFILE,
agentProfileRepoint,
@@ -2853,6 +2854,28 @@ async function cmdBrowser({ positional, flags }) {
timeout,
});
break;
+ case "key": {
+ const [action, key] = rest;
+ const example = "control-openhands browser key down Control";
+ if (action === "down") {
+ need(2, example);
+ let args;
+ try {
+ args = keyDownArguments(key, flags);
+ } catch (error) {
+ usage(error.message, error.hint);
+ }
+ result = await browserCall(run, "key-down", args);
+ } else if (action === "up") {
+ need(2, "control-openhands browser key up Control");
+ result = await browserCall(run, "key-up", { key });
+ } else if (action === "release-all") {
+ result = await browserCall(run, "key-release-all");
+ } else {
+ usage("browser key down|up <Key> | browser key release-all", example);
+ }
+ break;
+ }
case "select":
need(2, "control-openhands browser select 'testid=x' value");
result = await browserCall(run, "select", {
@@ -3367,6 +3390,7 @@ const BROWSER_VERBS = new Set([
"fill",
"type",
"press",
+ "key",
"select",
"upload",
"scroll",
@@ -4201,6 +4225,10 @@ Verbs
tooltip <sel> hover and return the role=tooltip text (or the title attribute)
fill|type <sel> [<value> | --value-file F | --value-env VAR] (prefer the file/env forms for keys)
press <Key> [--selector S] e.g. Escape, Enter, Control+k, Meta+k
+ key down <Key> [--repeat N [--interval-ms 30]] press and hold; N more keydowns with event.repeat set
+ key up <Key> | key release-all release one held key | all of them (last pressed first)
+ (held keys stay down across commands, and held modifiers apply
+ to later clicks and presses; browser reset or stop ends them)
select <sel> <value> | upload <sel> <file...> | scroll [<sel>] [--by PX] [--x PX]
[--observe SEL [--observe-ms 3000]] (scroll: record what SEL shows while the effects play out, e.g. a loading row)
upload-via <trigger-sel> <file...> click a button/menu item and answer the real file chooser
diff --git a/.agents/skills/verify-openhands/scripts/control-openhands.test.mjs b/.agents/skills/verify-openhands/scripts/control-openhands.test.mjs
index c5a7dd959b..c47c575a53 100755
--- a/.agents/skills/verify-openhands/scripts/control-openhands.test.mjs
+++ b/.agents/skills/verify-openhands/scripts/control-openhands.test.mjs
@@ -36,6 +36,7 @@ import { routePattern } from "./lib/route-pattern.mjs";
import { resolveTestids } from "./lib/testids.mjs";
import { tmuxPathFor } from "./lib/tmux-path.mjs";
import { browserCallLimit } from "./lib/call-limit.mjs";
+import { HeldKeys, keyDownArguments } from "./lib/held-keys.mjs";
import { agentProfileRepoint } from "./lib/agent-profile-repoint.mjs";
import {
canvasSize,
@@ -1621,6 +1622,141 @@ test.skipIf(!findFfmpeg())(
},
);
+// A page whose keyboard records calls stands in for Playwright's; the real
+// repeat flag is checked live (see the PR that added `browser key`).
+function keyPage(name, calls, { closed = false } = {}) {
+ return {
+ keyboard: {
+ async down(key) {
+ calls.push(`${name} down ${key}`);
+ },
+ async up(key) {
+ calls.push(`${name} up ${key}`);
+ if (closed)
+ throw new Error("Target page, context or browser has been closed");
+ },
+ },
+ };
+}
+
+test("held keys stay down, repeat on request and release where they were pressed", async () => {
+ const calls = [];
+ const first = keyPage("tab0", calls);
+ const second = keyPage("tab1", calls);
+ const keys = new HeldKeys();
+ assert.deepEqual(await keys.down(first, "Control"), {
+ key: "Control",
+ repeats: 0,
+ held: ["Control"],
+ });
+ const enter = await keys.down(first, "Enter", { repeat: 3, intervalMs: 0 });
+ assert.deepEqual(enter, {
+ key: "Enter",
+ repeats: 3,
+ held: ["Control", "Enter"],
+ });
+ // Down again on a held key is a repeat, on the tab it went down on.
+ await keys.down(second, "Enter", { intervalMs: 0 });
+ await keys.down(second, "Shift");
+ assert.deepEqual(await keys.up("Enter"), {
+ key: "Enter",
+ held: ["Control", "Shift"],
+ });
+ await assert.rejects(keys.up("Enter"), (error) => {
+ assert.equal(error.message, "Enter is not held");
+ assert.match(error.hint, /Held now: Control, Shift/);
+ return true;
+ });
+ assert.deepEqual(await keys.releaseAll(), {
+ released: ["Shift", "Control"],
+ held: [],
+ });
+ assert.deepEqual(calls, [
+ "tab0 down Control",
+ "tab0 down Enter",
+ "tab0 down Enter",
+ "tab0 down Enter",
+ "tab0 down Enter",
+ "tab0 down Enter",
+ "tab1 down Shift",
+ "tab0 up Enter",
+ "tab1 up Shift",
+ "tab0 up Control",
+ ]);
+ await assert.rejects(keys.up("Control"), (error) => {
+ assert.match(error.hint, /Nothing is held/);
+ return true;
+ });
+});
+
+test("releasing a key held on a closed tab still forgets it", async () => {
+ const calls = [];
+ const keys = new HeldKeys();
+ await keys.down(keyPage("gone", calls, { closed: true }), "Shift");
+ assert.deepEqual(await keys.releaseAll(), { released: ["Shift"], held: [] });
+ assert.deepEqual(keys.held, []);
+});
+
+test("key down flags are validated before the daemon is contacted", () => {
+ assert.deepEqual(keyDownArguments("Enter"), {
+ key: "Enter",
+ repeat: undefined,
+ intervalMs: undefined,
+ });
+ assert.deepEqual(
+ keyDownArguments("Enter", { repeat: "5", "interval-ms": "50" }),
+ {
+ key: "Enter",
+ repeat: 5,
+ intervalMs: 50,
+ },
+ );
+ for (const flags of [
+ { repeat: "-1" },
+ { repeat: "1.5" },
+ { repeat: "1001" },
+ { repeat: "x" },
+ { repeat: "2", "interval-ms": "-5" },
+ { "interval-ms": "50" },
+ { repeat: "1000", "interval-ms": "100" },
+ ]) {
+ assert.throws(
+ () => keyDownArguments("Enter", flags),
+ /--repeat|--interval-ms/,
+ );
+ }
+});
+
+test("browser help documents key, and bad key commands are usage errors", () => {
+ const help = spawnSync(process.execPath, [cli, "help", "browser"], {
+ encoding: "utf8",
+ });
+ assert.equal(help.status, 0);
+ assert.match(help.stdout, /key down <Key> \[--repeat N/);
+ assert.match(help.stdout, /key up <Key> \| key release-all/);
+ // A run directory gets the command past run lookup to its own checks;
+ // each case fails as a usage error before any browser is contacted.
+ const runDir = mkdtempSync(join(tmpdir(), "cov-key-"));
+ writeFileSync(join(runDir, "run.json"), "{}\n");
+ for (const args of [
+ ["key", "down"],
+ ["key", "hold", "Enter"],
+ ["key", "down", "Enter", "--repeat", "x"],
+ ["key", "down", "Enter", "--interval-ms", "50"],
+ ]) {
+ const bad = spawnSync(
+ process.execPath,
+ [cli, "browser", ...args, "--run", runDir],
+ { encoding: "utf8" },
+ );
+ assert.equal(
+ bad.status,
+ 2,
+ `${args.join(" ")}: ${bad.stdout}${bad.stderr}`,
+ );
+ }
+});
+
test("browser help documents record", () => {
const help = spawnSync(process.execPath, [cli, "help", "browser"], {
encoding: "utf8",
diff --git a/.agents/skills/verify-openhands/scripts/lib/held-keys.mjs b/.agents/skills/verify-openhands/scripts/lib/held-keys.mjs
new file mode 100644
index 0000000000..73ac7a0d58
--- /dev/null
+++ b/.agents/skills/verify-openhands/scripts/lib/held-keys.mjs
@@ -0,0 +1,117 @@
+// Keys held down across CLI calls (`browser key down|up|release-all`).
+//
+// Playwright's keyboard.down() on a key that is already down sends another
+// keydown with `event.repeat === true`, which is how a held key reaches the
+// page. A headless browser never auto-repeats, so repeats are explicit: a
+// count and an interval. Each held key remembers the page it went down on,
+// so switching tabs cannot strand it, and `key up` releases it there.
+// Modifiers held this way also apply to later clicks and presses.
+
+export const KEY_REPEAT_DEFAULTS = {
+ intervalMs: 30,
+ maxRepeat: 1000,
+ // Keeps one command well inside the CLI's browser call timeout.
+ maxTotalMs: 60_000,
+};
+
+const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+
+function keyError(message, hint) {
+ return Object.assign(new Error(message), { hint });
+}
+
+export class HeldKeys {
+ #keys = new Map();
+
+ /** Key names in the order they went down. */
+ get held() {
+ return [...this.#keys.keys()];
+ }
+
+ /**
+ * Press `key` and keep it down. A key that is already down gets repeat
+ * keydowns only: `repeat` of them, or one when no count is given.
+ */
+ async down(page, key, { repeat, intervalMs } = {}) {
+ const interval = intervalMs ?? KEY_REPEAT_DEFAULTS.intervalMs;
+ const target = this.#keys.get(key) ?? page;
+ let repeats = repeat ?? 0;
+ if (this.#keys.has(key)) {
+ repeats = repeat ?? 1;
+ } else {
+ await target.keyboard.down(key);
+ this.#keys.set(key, target);
+ }
+ for (let i = 0; i < repeats; i += 1) {
+ if (interval > 0) await sleep(interval);
+ await target.keyboard.down(key);
+ }
+ return { key, repeats, held: this.held };
+ }
+
+ async up(key) {
+ const page = this.#keys.get(key);
+ if (!page) {
+ throw keyError(
+ `${key} is not held`,
+ this.held.length
+ ? `Held now: ${this.held.join(", ")}`
+ : "Nothing is held; `browser key down <Key>` presses and holds a key.",
+ );
+ }
+ this.#keys.delete(key);
+ // A closed tab has already let go of its keys.
+ await page.keyboard.up(key).catch(() => {});
+ return { key, held: this.held };
+ }
+
+ /** Release every held key, last pressed first. */
+ async releaseAll() {
+ const released = this.held.reverse();
+ for (const key of released) await this.up(key);
+ return { released, held: this.held };
+ }
+}
+
+/** Validate CLI flags for `browser key down`; returns daemon arguments. */
+export function keyDownArguments(key, flags = {}) {
+ const repeat = flags.repeat === undefined ? undefined : Number(flags.repeat);
+ if (
+ repeat !== undefined &&
+ (!Number.isInteger(repeat) ||
+ repeat < 0 ||
+ repeat > KEY_REPEAT_DEFAULTS.maxRepeat)
+ ) {
+ throw keyError(
+ `--repeat must be a whole number from 0 to ${KEY_REPEAT_DEFAULTS.maxRepeat}`,
+ "control-openhands browser key down Enter --repeat 5 --interval-ms 50",
+ );
+ }
+ const intervalMs =
+ flags["interval-ms"] === undefined
+ ? undefined
+ : Number(flags["interval-ms"]);
+ if (
+ intervalMs !== undefined &&
+ (!Number.isFinite(intervalMs) || intervalMs < 0 || intervalMs > 10_000)
+ ) {
+ throw keyError(
+ "--interval-ms must be a number of milliseconds from 0 to 10000",
+ "control-openhands browser key down Enter --repeat 5 --interval-ms 50",
+ );
+ }
+ if (intervalMs !== undefined && repeat === undefined) {
+ throw keyError(
+ "--interval-ms only applies with --repeat",
+ "control-openhands browser key down Enter --repeat 5 --interval-ms 50",
+ );
+ }
+ const total = (repeat ?? 0) * (intervalMs ?? KEY_REPEAT_DEFAULTS.intervalMs);
+ if (total > KEY_REPEAT_DEFAULTS.maxTotalMs) {
+ throw keyError(
+ `--repeat times --interval-ms is ${total} ms; keep it within ${KEY_REPEAT_DEFAULTS.maxTotalMs} ms`,
+ "Split a long hold into several `browser key down <Key> --repeat N` calls.",
+ );
+ }
+ return { key, repeat, intervalMs };
+}
diff --git a/.pr/bugs/2026-10-10/draft-run-status/first/completed-after-reload.png b/.pr/bugs/2026-10-10/draft-run-status/first/completed-after-reload.png
deleted file mode 100644
index fbefbabe1a..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/first/completed-after-reload.png and /dev/null differ
diff --git a/.pr/bugs/2026-10-10/draft-run-status/first/custom-script-test.gif b/.pr/bugs/2026-10-10/draft-run-status/first/custom-script-test.gif
deleted file mode 100644
index a1520856e5..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/first/custom-script-test.gif and /dev/null differ
diff --git a/.pr/bugs/2026-10-10/draft-run-status/first/custom-script-test.mp4 b/.pr/bugs/2026-10-10/draft-run-status/first/custom-script-test.mp4
deleted file mode 100644
index 364fda894e..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/first/custom-script-test.mp4 and /dev/null differ
diff --git a/.pr/bugs/2026-10-10/draft-run-status/first/sanitized-run-summary.json b/.pr/bugs/2026-10-10/draft-run-status/first/sanitized-run-summary.json
deleted file mode 100644
index df2238bdf6..0000000000
--- a/.pr/bugs/2026-10-10/draft-run-status/first/sanitized-run-summary.json
+++ /dev/null
@@ -1,18 +0,0 @@
-{
- "revision": "5a8c3832154b14d2dad651801c0ab3f74c96f02e",
- "agent_server": "1.54.0",
- "automation": "1.19.3",
- "viewport": "1440x1000",
- "backend": "local; no-conversation; generated callback-only custom script",
- "run_id": "3aff8e91-c7c3-4e02-8a52-bcf8025b5265",
- "automation_id": "5a1f553d-2a16-4cb2-8464-aad7f6822489",
- "status": "COMPLETED",
- "created_at": "2026-10-10T07:24:20Z",
- "started_at": "2026-10-10T07:24:27.818238Z",
- "completed_at": "2026-10-10T07:24:27.939441Z",
- "ui_before_reload": "Pending for at least 45 seconds after confirmed backend completion",
- "ui_after_reload": "Successful; draft form Latest test passed",
- "origin": "unconfirmed; no baseline live proof",
- "model_calls": 0,
- "cost_usd": 0
-}
diff --git a/.pr/bugs/2026-10-10/draft-run-status/first/stale-pending-45s.png b/.pr/bugs/2026-10-10/draft-run-status/first/stale-pending-45s.png
deleted file mode 100644
index 1be4e6873c..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/first/stale-pending-45s.png and /dev/null differ
diff --git a/.pr/bugs/2026-10-10/draft-run-status/second/second-completed-after-reload.png b/.pr/bugs/2026-10-10/draft-run-status/second/second-completed-after-reload.png
deleted file mode 100644
index 868482de83..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/second/second-completed-after-reload.png and /dev/null differ
diff --git a/.pr/bugs/2026-10-10/draft-run-status/second/second-sanitized-run-summary.json b/.pr/bugs/2026-10-10/draft-run-status/second/second-sanitized-run-summary.json
deleted file mode 100644
index 9ac9f0be9b..0000000000
--- a/.pr/bugs/2026-10-10/draft-run-status/second/second-sanitized-run-summary.json
+++ /dev/null
@@ -1,27 +0,0 @@
-{
- "revision": "5a8c3832154b14d2dad651801c0ab3f74c96f02e",
- "agentServerVersion": "1.54.0",
- "automationVersion": "1.19.3",
- "automationVersionExplicitFlag": true,
- "viewport": "1440x1000",
- "selectedBackend": "Local",
- "scenario": "No conversation; generated custom callback-only script; no LLM key/profile/model call",
- "draftId": "b319eaa8-4175-4bd2-ad03-9e8ba1a60e91",
- "automationId": "607fbc9b-a3fb-48c9-8d6b-ef6f498ba4ca",
- "runId": "94393a46-dcf9-4292-9955-ea7bb66eeb2b",
- "backendStatusBeforeWait": "COMPLETED",
- "backendStatusAfterWait": "COMPLETED",
- "runCreatedAt": "2026-10-10T07:36:47Z",
- "runStartedAt": "2026-10-10T07:36:49.894555Z",
- "runCompletedAt": "2026-10-10T07:36:50.016567Z",
- "successfulRowWaitStartedAt": "2026-10-10T07:37:16.148575+00:00",
- "successfulRowWaitEndedAt": "2026-10-10T07:38:01.306675+00:00",
- "waitSeconds": 45.1581,
- "successfulRowWaitTimedOut": true,
- "visibleRowBeforeReload": "Pending",
- "visibleRowAfterReload": "Successful",
- "doctorAfterStaleUI": "ok",
- "origin": "observed-origin-unconfirmed; no baseline run",
- "modelCalls": 0,
- "modelCostUSD": 0
-}
diff --git a/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-pending-45s.png b/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-pending-45s.png
deleted file mode 100644
index 8bb2f1036b..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-pending-45s.png and /dev/null differ
diff --git a/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-test-runs.gif b/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-test-runs.gif
deleted file mode 100644
index 1a35da9bc5..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-test-runs.gif and /dev/null differ
diff --git a/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-test-runs.mp4 b/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-test-runs.mp4
deleted file mode 100644
index fe86d87208..0000000000
Binary files a/.pr/bugs/2026-10-10/draft-run-status/second/second-stale-test-runs.mp4 and /dev/null differ
diff --git a/.pr/bugs/mcp-header-error-after.png b/.pr/bugs/mcp-header-error-after.png
deleted file mode 100644
index 6cb07cb402..0000000000
Binary files a/.pr/bugs/mcp-header-error-after.png and /dev/null differ
diff --git a/.pr/bugs/mcp-header-error-before.png b/.pr/bugs/mcp-header-error-before.png
deleted file mode 100644
index a6a19e2ab0..0000000000
Binary files a/.pr/bugs/mcp-header-error-before.png and /dev/null differ
diff --git a/.pr/bugs/skill-modal-pill-escape-after.png b/.pr/bugs/skill-modal-pill-escape-after.png
deleted file mode 100644
index a33778a936..0000000000
Binary files a/.pr/bugs/skill-modal-pill-escape-after.png and /dev/null differ
diff --git a/.pr/bugs/skill-modal-pill-escape-before.png b/.pr/bugs/skill-modal-pill-escape-before.png
deleted file mode 100644
index 52154dbee2..0000000000
Binary files a/.pr/bugs/skill-modal-pill-escape-before.png and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/deepseek-cloud-receipts.json b/.pr/map-maintenance/2026-10-10/deepseek-cloud-receipts.json
deleted file mode 100644
index d8f54620fe..0000000000
--- a/.pr/map-maintenance/2026-10-10/deepseek-cloud-receipts.json
+++ /dev/null
@@ -1,522 +0,0 @@
-{
- "provenance": {
- "kind": "filtered_transcription_of_completed_cloud_command_outputs",
- "task_id": "01a1257b-9287-7597-94ac-0dc8ba9e606e",
- "task_title": "Fix Codex DeepSeek Key Usage",
- "turn_id": "01a1259a-f52c-7603-92cb-0e3c2e45faf4",
- "test_date_utc": "2026-10-10",
- "reconciled_date": "2026-10-11",
- "source_command_ids": [
- "exec-f0980313-6ff6-4e7c-a5ce-6b87a7a3942d",
- "exec-556abb5f-f12a-48d7-8bea-4bb7ca47aa4f",
- "exec-d92d60d8-00d0-4637-82bf-ebcf6f543d85",
- "exec-b06da630-6d93-4401-8e31-78e0f775677a",
- "exec-3227f861-006a-4026-aece-4946467e250a",
- "exec-7ee8a4dc-ec8e-47c8-ac52-491c49862f6b"
- ],
- "original_ledgers_and_media_copied": false
- },
- "identity": {
- "checkout_revision": "ac3ea58e42d09178f027878a66b7b2e36bf59abd",
- "product_revision": "5a8c3832154b14d2dad651801c0ab3f74c96f02e",
- "accepted_base": "553d4519113192a80fb18376fdb6c1e39fa06387",
- "runtime": {
- "backend": "Local",
- "AgentServer": "1.54.0",
- "SDK": "1.54.0",
- "automation": "1.19.3",
- "Node": "v24.19.0",
- "Chromium": "151.0.7922.173",
- "desktop": "1440x1000",
- "phone": "390x844",
- "buildIdentity": "31b4983403910d82",
- "OS": "Linux",
- "execution_host": "OpenAI Codex cloud",
- "selected_openhands_backend": "Local",
- "model": "deepseek/deepseek-flash"
- }
- },
- "latest_scoped_check_counts": {
- "pass": 40,
- "fail": 4,
- "blocked": 1
- },
- "groups": [
- {
- "assignment": "assisted-forms",
- "latest_scoped_checks": 13,
- "results": {
- "pass": 12,
- "fail": 1
- },
- "source_ledger_sha256_reported_by_root": "05c791207ddfa00648f5be6541814713eb5316ba7a3d9176498e4fb6bc26a1ad"
- },
- {
- "assignment": "execution-and-outcomes",
- "latest_scoped_checks": 27,
- "results": {
- "pass": 23,
- "fail": 3,
- "blocked": 1
- },
- "source_ledger_sha256_reported_by_root": "b879b406d808662802a247b45074083fd3e4ebb046b8665e372e838d69bdc4ca"
- },
- {
- "assignment": "long-finish-logs",
- "latest_scoped_checks": 5,
- "results": {
- "pass": 5
- },
- "source_ledger_sha256_reported_by_root": "d11374ee95fa27b245f66c9dbefd86abcc9045cdf8b422e8dea33a1ea8561af5"
- }
- ],
- "assisted_forms": {
- "producers": [
- {
- "kind": "prompt",
- "entry_viewport": "desktop1440x1000",
- "status": "finished",
- "model": "deepseek/deepseek-flash",
- "action_tools": [
- "automation_form_update"
- ],
- "SDK_recorded_cost_USD": 0.003985674
- },
- {
- "kind": "prompt",
- "entry_viewport": "phone390x844",
- "status": "finished",
- "model": "deepseek/deepseek-flash",
- "action_tools": [
- "automation_form_update"
- ],
- "SDK_recorded_cost_USD": 0.0038454779999999994
- },
- {
- "kind": "plugin",
- "entry_viewport": "desktop1440x1000",
- "status": "finished",
- "model": "deepseek/deepseek-flash",
- "action_tools": [
- "automation_form_update"
- ],
- "SDK_recorded_cost_USD": 0.0039015119999999998
- },
- {
- "kind": "plugin",
- "entry_viewport": "phone390x844",
- "status": "finished",
- "model": "deepseek/deepseek-flash",
- "action_tools": [
- "automation_form_update"
- ],
- "SDK_recorded_cost_USD": 0.003913811999999999
- }
- ],
- "executions": [
- {
- "kind": "agent-created prompt Run now",
- "entry_viewport": "phone390x844",
- "completed_at": "2026-10-10T11:50:21.619455Z",
- "automatic_detail_update": "Successful without reload",
- "status": "COMPLETED",
- "model": "deepseek/deepseek-flash",
- "agent_message": "pong",
- "action_tools": [
- "finish"
- ],
- "SDK_recorded_cost_USD": 0.0009070739999999999
- },
- {
- "kind": "agent-created plugin Run now",
- "entry_viewport": "desktop1440x1000",
- "completed_at": "2026-10-10T11:52:17.358209Z",
- "automatic_detail_update": "Successful without reload",
- "status": "COMPLETED",
- "model": "deepseek/deepseek-flash",
- "agent_message": "pong",
- "action_tools": [
- "finish"
- ],
- "SDK_recorded_cost_USD": 0.000966714
- },
- {
- "kind": "agent-created plugin draft Test",
- "entry_viewport": "phone390x844",
- "completed_at": "2026-10-10T11:53:49.600611Z",
- "automatic_detail_update": "FAIL: open Test runs remains Pending; reload recovery succeeds",
- "status": "COMPLETED",
- "model": "deepseek/deepseek-flash",
- "agent_message": "pong",
- "action_tools": [
- "finish"
- ],
- "SDK_recorded_cost_USD": 0.000981738
- },
- {
- "kind": "agent-created plugin final Run now after draft Test",
- "entry_viewport": "phone390x844",
- "completed_at": "2026-10-10T11:58:32.958930Z",
- "automatic_detail_update": "Successful without reload",
- "status": "COMPLETED",
- "model": "deepseek/deepseek-flash",
- "agent_message": "pong",
- "action_tools": [
- "finish"
- ],
- "SDK_recorded_cost_USD": 0.0008962499999999999
- }
- ],
- "latest_entries": [
- {
- "entry": "Local desktop: Add Automation > Create > docked prompt",
- "result": "pass",
- "viewport": "1440x1000"
- },
- {
- "entry": "Local desktop: Hide Agent > Show Agent",
- "result": "pass",
- "viewport": "1440x1000"
- },
- {
- "entry": "Local phone: existing setup Form > Conversation",
- "result": "pass",
- "viewport": "390x844"
- },
- {
- "entry": "Local phone: dashboard Resume draft > reload",
- "result": "pass",
- "viewport": "390x844"
- },
- {
- "entry": "Local phone: associated conversation > Form",
- "result": "pass",
- "viewport": "390x844"
- },
- {
- "entry": "Local desktop: assisted form > Create Automation",
- "result": "pass",
- "viewport": "1440x1000"
- },
- {
- "entry": "Local phone: fresh Add Automation > docked prompt > Save > reload > Create",
- "result": "pass",
- "viewport": "390x844"
- },
- {
- "entry": "Local phone: agent-created prompt > Turn on > Run now",
- "result": "pass",
- "viewport": "390x844"
- },
- {
- "entry": "Local desktop: genuine plugin form > Create > Turn on > Run now",
- "result": "pass",
- "viewport": "1440x1000"
- },
- {
- "entry": "Local phone: real plugin draft Test > automatic status update",
- "result": "fail",
- "viewport": "390x844"
- },
- {
- "entry": "Local phone: fresh genuine plugin form > Save > Create > Turn on > Run now",
- "result": "pass",
- "viewport": "390x844"
- },
- {
- "entry": "Local desktop: Templates > Custom automation > Prompt > flash > Review > Back > Confirm",
- "result": "pass",
- "viewport": "1440x1000"
- },
- {
- "entry": "Local phone: real plugin draft Test > terminal backend > reload readback > Back",
- "result": "pass",
- "viewport": "390x844"
- }
- ],
- "draft_test_failure": {
- "issue": "https://github.com/OpenHands/OpenHands/issues/18263",
- "backendStatus": "COMPLETED",
- "backendCompletedAt": "2026-10-10T11:53:49.600611Z",
- "staleScreenshotAt": "2026-10-10T11:57:10.962621Z",
- "staleAfterCompletionSeconds": 201.36201,
- "UI": "Pending without reload",
- "recovery": "Reload then View test runs showed Successful, pong outcome, $0.0010; Back showed Latest test passed."
- },
- "plugin_loading": {
- "registration_confirmed_in_execution_conversations": 3,
- "skill_name": "magic-word",
- "source": "public packaged magic-test plugin",
- "skill_triggers_exercised": false,
- "hooks_MCP_scripts_exercised": false
- },
- "costs": {
- "producerUSD": 0.015646476,
- "executionUSD": 0.0037517759999999996,
- "totalSDKRecordedUSD": 0.019398252,
- "SDKRecordedModelRequestCount": 16,
- "scope": "Eight owned conversation SDK stats; excludes preset/key validation and unrecorded helper/title calls. Costs are SDK estimates, not provider billing receipts. No prompt budget cap imposed."
- },
- "uncovered": [
- "Phone template Review/Back/Confirm.",
- "Distinct template Plugin action/review/timezone path; agent-assisted plugin creation and execution passed on desktop and phone.",
- "Plugin magic-word trigger/invoke_skill behavior; pong avoids these triggers.",
- "Remote/cloud OpenHands backend, external integrations, other models, complete F22 family."
- ]
- },
- "execution_and_outcomes": {
- "distinct_feature_ids": [
- "F22.agent-assisted-setup",
- "F22.setup-review-confirm",
- "F23.activity-log",
- "F23.activity-log-export",
- "F23.debug-with-openhands",
- "F23.run-linked-conversation",
- "F23.run-logs-modal",
- "F23.run-no-conversation",
- "F23.run-now",
- "F23.run-status-polling",
- "F23.run-task-outcome"
- ],
- "completed_runs": [
- {
- "scenario": "Desktop Pong",
- "outcome": "success",
- "SDK_reported_cost_USD": 0.000977364
- },
- {
- "scenario": "Phone Pong",
- "outcome": "success",
- "SDK_reported_cost_USD": 0.000887274
- },
- {
- "scenario": "Blocked",
- "outcome": "blocked",
- "SDK_reported_cost_USD": 0.00079038
- },
- {
- "scenario": "Partial",
- "outcome": "partial_success",
- "SDK_reported_cost_USD": 0.00079143
- },
- {
- "scenario": "Unknown",
- "outcome": "unknown",
- "SDK_reported_cost_USD": 0.000526206
- },
- {
- "scenario": "Independent prompt draft Test",
- "outcome": "success",
- "SDK_reported_cost_USD": 0.0008745
- }
- ],
- "pass_entries": [
- {
- "feature": "F22.setup-review-confirm",
- "entry": "Direct /automations/new/custom-automation; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-now",
- "entry": "Run now from F22-created detail; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-status-polling",
- "entry": "Run now detail stays loaded; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.activity-log",
- "entry": "Empty→genuine completed run; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-linked-conversation",
- "entry": "Activity row link then Browser Back; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.activity-log-export",
- "entry": "Activity Log Export JSON and CSV; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-logs-modal",
- "entry": "View logs for genuine Pong; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F22.setup-review-confirm",
- "entry": "Templates card→custom-automation; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-now",
- "entry": "Run now from independently phone-created detail; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-status-polling",
- "entry": "Phone Run now detail staysloaded; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.activity-log",
- "entry": "Genuine phone-created prompt run; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-linked-conversation",
- "entry": "Activity row link and Browser Back; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.activity-log-export",
- "entry": "Activity Export JSON/CSV; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-logs-modal",
- "entry": "View logs phone Pong; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-task-outcome",
- "entry": "Genuine finish blocked; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-task-outcome",
- "entry": "Genuine finish partial; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-task-outcome",
- "entry": "Genuine finish unknown; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.debug-with-openhands",
- "entry": "Genuine Failed prompt run Logs→Debug; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-task-outcome",
- "entry": "Genuine finish blocked; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-task-outcome",
- "entry": "Genuine finish partial; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-task-outcome",
- "entry": "Genuine finish unknown; desktop 1440x1000; selected Local default-local"
- },
- {
- "feature": "F23.run-logs-modal",
- "entry": "Genuine failed timeoutprompt Logs; phone 390x844; selected Local default-local"
- },
- {
- "feature": "F23.run-logs-modal",
- "entry": "Genuine failed timeoutprompt Logs; desktop 1440x1000; selected Local default-local"
- }
- ],
- "failures": [
- {
- "feature": "F22.agent-assisted-setup",
- "known_issue": "Canvas#18263",
- "scope": "Desktop genuine prompt draft Test stays Pending after backend COMPLETED and45.263seconds successfulrow wait; reload showsSuccessful",
- "proof": "F22.agent-assisted-setup/independent-genuine-status-proof.json"
- },
- {
- "feature": "F23.run-no-conversation",
- "known_issue": "Canvas#17948",
- "viewports": [
- "1440x1000",
- "390x844"
- ],
- "scope": "Failed promptpreset withoutconversation incorrectly labelled script"
- }
- ],
- "failed_producer": {
- "status": "FAILED",
- "conversation_present": false,
- "reason": "Timed out: command timed out or was killed",
- "timeout_seconds": 1
- },
- "debug_launch": {
- "status_at_cleanup": "paused",
- "model": "deepseek/deepseek-flash",
- "reported_metrics_cost_USD": 0.005270172,
- "observed_actions": 5,
- "scope": "Only launch, correct seed/skill and initial agent execution verified; no repair completion claimed. Own workspace and own fixture API GETs; no automation writes before pause."
- },
- "remaining_prerequisites": [
- "Genuine legacy/custom response-schema finish producer for absent/nonstring/unrecognized status fallback; genuine failed producer carrying finish summary for summary-retention boundary."
- ],
- "costs": {
- "automation_run_reported_cost_total": 0.004847154,
- "cost_limit_enforced_by_this_verification": false,
- "reported_cost_caveat": "Reported automation-run costs only; model validation and title generation are not claimed to be fully accounted here."
- }
- },
- "long_finish_logs": {
- "run_id": "f23-genuine-finish",
- "UTC_interval": [
- "2026-10-10T11:43:41.751Z",
- "2026-10-10T11:53:05.822Z"
- ],
- "genuine_finish": {
- "model": "deepseek/deepseek-flash",
- "status": "success",
- "outcome_summary_characters": 4596,
- "metadata_message_characters": 1079,
- "model_actions": [
- "finish"
- ],
- "metadata_injected": false,
- "LLM_mocked": false
- },
- "viewport_checks": [
- {
- "feature": "F23.run-logs-modal",
- "viewport": "1440x1000",
- "result": "pass",
- "observed": "Summary/metadata endings, Output/Error tabs and true stdout end reachable; sticky header/X visible; closes without navigation; no phone horizontal overflow."
- },
- {
- "feature": "F23.run-logs-modal",
- "viewport": "390x844",
- "result": "pass",
- "observed": "Summary/metadata endings, Output/Error tabs and true stdout end reachable; sticky header/X visible; closes without navigation; no phone horizontal overflow."
- },
- {
- "feature": "F23.run-logs-modal",
- "viewport": "1440x500",
- "result": "pass",
- "observed": "Summary/metadata endings, Output/Error tabs and true stdout end reachable; sticky header/X visible; closes without navigation; no phone horizontal overflow."
- },
- {
- "feature": "F23.run-logs-modal",
- "viewport": "390x500",
- "result": "pass",
- "observed": "Summary/metadata endings, Output/Error tabs and true stdout end reachable; sticky header/X visible; closes without navigation; no phone horizontal overflow."
- }
- ],
- "desktop_completion_polling": {
- "feature": "F23.run-status-polling",
- "viewport": "1440x1000",
- "result": "pass",
- "observed": "Pending/Running became Successful without reload; summary, rounded cost and conversation link appeared; backend COMPLETED."
- },
- "recorded_costs_USD": {
- "long_producer": 0.001988394,
- "all_four_producers": 0.004413984,
- "other_three_short_producers_excluded_from_coverage": true
- }
- },
- "cleanup": {
- "remaining_fixture_automations": 0,
- "remaining_fixture_drafts": 0,
- "all_test_ports_reported_closed": true,
- "live_owned_group_processes": 0,
- "dead_unreaped_zombie_PIDs_caused_cli_stop_false": true,
- "repository_trees_reported_clean": true,
- "evidence_preserved": true,
- "final_reported_text_secret_scan": {
- "files_scanned": 384,
- "credential_values_checked": 2,
- "matching_paths": []
- }
- },
- "limits": {
- "full_delta_complete": false,
- "baseline_advanced": false,
- "OpenHands_Cloud_backend_verified": false,
- "complete_provider_billing_or_aggregate_budget_enforcement_proven": false,
- "local_followup_independently_inspected_cloud_images": false
- }
-}
diff --git a/.pr/map-maintenance/2026-10-10/deepseek-cloud-results.md b/.pr/map-maintenance/2026-10-10/deepseek-cloud-results.md
deleted file mode 100644
index 47526e425e..0000000000
--- a/.pr/map-maintenance/2026-10-10/deepseek-cloud-results.md
+++ /dev/null
@@ -1,65 +0,0 @@
-I'm an AI agent (Codex) helping Engel Nyst (@enyst) with project work.
-
-# Verified DeepSeek scopes: 40 pass, 4 fail, 1 producer block
-
-Verified on 2026-10-10 in the completed OpenAI Codex cloud session **Fix Codex DeepSeek Key Usage**, task `01a1257b-9287-7597-94ac-0dc8ba9e606e`, verification turn `01a1259a-f52c-7603-92cb-0e3c2e45faf4`; incorporated into this PR on 2026-10-11. Three independent agents used fresh, separate stacks and genuine `deepseek/deepseek-flash` calls. Their final reports and command outputs supply the scope and results below.
-
-The 45 reported entry/check scopes cover parts of F22 and F23. **40 passed, four failed on two existing defects, and one remains blocked on a matching legacy/custom-schema producer.** They are not 45 distinct feature IDs, a complete family pass, or a completed 745-ID delta. Counts apply to the three assignments below; overlapping credential-free checks are reported separately.
-
-## Revision and runtime
-
-- Tested checkout: `ac3ea58e42d09178f027878a66b7b2e36bf59abd`, the documentation-only head of [PR #18264](https://github.com/OpenHands/OpenHands/pull/18264). Product source matches frozen TARGET `5a8c3832154b14d2dad651801c0ab3f74c96f02e`.
-- Accepted BASE remains `553d4519113192a80fb18376fdb6c1e39fa06387`.
-- OpenAI cloud Linux; the selected **OpenHands backend was Local**, not OpenHands Cloud. Canvas 1.26.0, Agent Server/SDK 1.54.0, automation 1.19.3; build identity `31b4983403910d82`.
-- Node 24.19.0, npm 11.9.0 and Chromium 151.0.7922.173. Desktop 1440×1000, phone 390×844; long-log checks additionally used 1440×500 and 390×500. Doctor checks passed. No model output or run metadata was injected.
-
-## Separate agent scopes
-
-| Agent assignment | Entry/check scopes | Pass | Fail | Blocked |
-|---|---:|---:|---:|---:|
-| Assisted forms (`F22.agent-assisted-setup`, `F22.setup-review-confirm`) | 13 | 12 | 1 | 0 |
-| Execution and outcomes (F22/F23) | 27 | 23 | 3 | 1 |
-| Genuine long finish logs and desktop completion polling | 5 | 5 | 0 | 0 |
-| Total | 45 | 40 | 4 | 1 |
-
-The final agent summaries and source-ledger SHA-256 hashes are retained in [transcribed receipts](deepseek-cloud-receipts.json). This is a filtered transcription of completed cloud command outputs, not a replacement for the original append-only ledgers or media. The originating session retained those private artifacts and reports that its root visually inspected the passing and failing captures. This local follow-up did not download or independently inspect those images, and publishes no fabricated media or inaccessible screenshot links.
-
-## Verified behavior
-
-| Stable feature / entry point | Expected and observed result | Scope |
-|---|---|---|
-| `F22.agent-assisted-setup`: fresh Add Automation → Create → docked prompt | Genuine `automation_form_update` filled prompt and plugin forms; Save/reload/resume preserved values; Form/Conversation and Hide/Show preserved the form | Desktop and phone, selected Local |
-| `F22.agent-assisted-setup`: create → enable → Run now | Agent-created prompt/plugin automations completed with real `finish` events and became Successful without reloading the ordinary detail page | Prompt on phone; plugin on desktop and phone |
-| `F22.setup-review-confirm`: template/direct custom setup → Prompt → create | Desktop Review/Back/Confirm retained values; separate desktop and phone prompt creation used the selected flash profile | Phone Back/review persistence and the distinct template Plugin review/timezone path remain separate checks |
-| `F23.run-now`, `F23.run-status-polling`, `F23.activity-log` | Dispatch and live completion, summary, cost and conversation link appeared; backend independently reported COMPLETED | Desktop and phone |
-| `F23.run-linked-conversation`, `F23.activity-log-export` | Conversation link/Back worked; real JSON and CSV activity exports contained the run data | Desktop and phone |
-| `F23.run-task-outcome` | Genuine `blocked`, `partial_success` and `unknown` finishes rendered Blocked, Partial and Needs review respectively | Desktop and phone; normal TaskOutcome schema only |
-| `F23.debug-with-openhands`: failed run → Logs → Debug | Correct failed-run seed and automation skill reached a real DeepSeek conversation; initial execution observed, then paused | Desktop; launch/handoff only, no completed repair claim |
-| `F23.run-logs-modal`: completed/failed ordinary runs | Run/task/system sections, stdout/stderr tabs and closing worked | Desktop and phone |
-| `F23.run-logs-modal`: genuine long summary and metadata | Summary/metadata endings, Output/Error tabs and true stdout end reachable; sticky header/X visible; closing did not navigate; phone had no horizontal overflow | 1440×1000, 390×844, 1440×500, 390×500 |
-
-The plugin fixture was a copy of the public packaged `magic-test` plugin. Retained SystemPromptEvent checks confirmed registration of its `magic-word` skill in three execution conversations. The `pong` request did not invoke that skill's trigger, hooks, MCP or scripts; do not treat plugin loading/execution as proof of those additional behaviors.
-
-The long-log producer supplied a **4,596-character outcome_summary and 1,079-character metadata message**, with only a `finish` model action. No backend metadata was edited. Four viewport scopes plus genuine desktop status polling passed, verifying [#18173](https://github.com/OpenHands/OpenHands/issues/18173) / [fix #18175](https://github.com/OpenHands/OpenHands/pull/18175) on pinned Linux Local. F23 now states the passing long-content contract. MacOS/OpenHands Cloud need their own checks; no pre-fix live comparison is claimed.
-
-## Existing failures and exact remaining prerequisite
-
-- **[#18263](https://github.com/OpenHands/OpenHands/issues/18263), two failed scopes:** genuine phone plugin draft Test and an independently produced desktop prompt draft Test finished in the backend, but the open Test runs page stayed Pending. The phone capture was 201.362 seconds after backend completion. The desktop agent confirmed COMPLETED before and after a 45.263-second Successful-row wait. Reload recovered Successful in both cases. The phone capture was at 11:57:10.962621Z following completion at 11:53:49.600611Z.
-- **[#17948](https://github.com/OpenHands/OpenHands/issues/17948), two failed scopes:** an actual timeout-1 prompt-preset run ended FAILED without a conversation. Desktop and phone incorrectly called it a script run. This verifies the labeling defect, not every other symptom bundled into that issue.
-- **`F23.run-task-outcome`, one blocked scope:** a genuine legacy/custom-schema finish producer is needed for absent/non-string/unrecognized status fallback, plus an actual failed producer carrying a finish summary for failed-run summary retention. The normal TaskOutcome producer verifies only the supported enum statuses. The block is not a missing DeepSeek key or a budget failure; injected metadata and mocked responses are not substitutes.
-
-No duplicate issues were filed. No causal regression or complete product repair is claimed. These failures remain documented as failures.
-
-## Credentials, cost and coverage limits
-
-The supplied DeepSeek key passed model listing, a minimal real flash completion, profile validation and product execution. Credential-backed form assistance and automation execution are verified in the runtime described above.
-
-Recorded SDK estimates were $0.019398252 for eight assisted-form/execution conversations, $0.004847154 for six execution/outcome runs, $0.005270172 for the debug conversation, and $0.004413984 for the long-log worker's four actual producers. Three short producers in the last amount were excluded from coverage after overlapping another assignment. These are partial, differently scoped receipts, not a provider billing total: profile/preset validation, title generation and unrecorded helper calls are excluded. The verification did not enforce or prove the scheduled automation's $10 aggregate cap. Working product checks must not be represented as proof of comprehensive budget enforcement.
-
-Full F22/F23 coverage, distinct template Plugin review/timezone paths, plugin skill invocation, external integrations, other model profiles, OpenHands Cloud, native/Docker/embedded paths and the rest of the affected map remain outside this follow-up. Preserve the remaining prerequisites and keep BASE unchanged.
-
-## Teardown and changes made
-
-Final receipts report zero remaining fixture automations/drafts, every test port closed, no live process in the owned launcher groups, and model conversations finished or paused. Dead unreaped Linux zombie PIDs caused CLI `stop` to report `launcherStopped=false`; independent process-state and port checks confirmed shutdown. Evidence survived teardown. The originating root scanned 384 evidence text files against its two configured credential values with zero matches; raw prompts, private homes, keys, browser state and logs remain unpublished.
-
-The maintained F22 recipes incorporate the exercised form, navigation, creation and execution commands; the index identifies those verified paths, and F23 states the passing long-log behavior. BASE remains unchanged while the full delta is incomplete.
diff --git a/.pr/map-maintenance/2026-10-10/draft-peer/fresh-phone-draft.png b/.pr/map-maintenance/2026-10-10/draft-peer/fresh-phone-draft.png
deleted file mode 100644
index f90adbd4ec..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/draft-peer/fresh-phone-draft.png and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/draft-root/draft-desktop.mp4 b/.pr/map-maintenance/2026-10-10/draft-root/draft-desktop.mp4
deleted file mode 100644
index bfb364de8c..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/draft-root/draft-desktop.mp4 and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/draft-root/resume-phone.mp4 b/.pr/map-maintenance/2026-10-10/draft-root/resume-phone.mp4
deleted file mode 100644
index 4bc2106fcd..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/draft-root/resume-phone.mp4 and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/draft-root/resumed-phone.png b/.pr/map-maintenance/2026-10-10/draft-root/resumed-phone.png
deleted file mode 100644
index 3e6de1f902..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/draft-root/resumed-phone.png and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-export-results.json b/.pr/map-maintenance/2026-10-10/encryption-root/desktop-export-results.json
deleted file mode 100644
index 3b248b756b..0000000000
--- a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-export-results.json
+++ /dev/null
@@ -1,58 +0,0 @@
-{
- "viewport": "desktop",
- "selectedBackend": "Local",
- "automationVersion": "1.19.3",
- "results": [
- {
- "action": "key-only set",
- "dirtyCount": 1,
- "before": {
- "sha": "6035ce5a57c0a0dc67c602e37c86edf61b079950",
- "sha256": "bc4ad7a6280091680f7a00e6d83fac1b9bbc762d87fcc339b439013978d2fe36",
- "prefix": "agent_profile_id: nu",
- "encrypted": false
- },
- "after": {
- "sha": "c3523b145f00e802205d2116fd7f4d57b866a07c",
- "sha256": "a34a8de9e291a53b349416163e28bb575357d23a856f6bcfa5308474d6346f76",
- "prefix": "gAAAAABqyedTAbu4IiEw",
- "encrypted": true
- },
- "pass": true
- },
- {
- "action": "key-only rotate",
- "dirtyCount": 1,
- "before": {
- "sha": "c3523b145f00e802205d2116fd7f4d57b866a07c",
- "sha256": "a34a8de9e291a53b349416163e28bb575357d23a856f6bcfa5308474d6346f76",
- "prefix": "gAAAAABqyedTAbu4IiEw",
- "encrypted": true
- },
- "after": {
- "sha": "0b7c56de51441cf026c16a85d55e398f2e8f1e3f",
- "sha256": "6f70a1244ad84e095b0366ac4feb7f4a652e5d0de32a19f0e6f85a95f257370c",
- "prefix": "gAAAAABqyedYk_e4lONk",
- "encrypted": true
- },
- "pass": true
- },
- {
- "action": "key-only clear",
- "dirtyCount": 1,
- "before": {
- "sha": "0b7c56de51441cf026c16a85d55e398f2e8f1e3f",
- "sha256": "6f70a1244ad84e095b0366ac4feb7f4a652e5d0de32a19f0e6f85a95f257370c",
- "prefix": "gAAAAABqyedYk_e4lONk",
- "encrypted": true
- },
- "after": {
- "sha": "dc1941b3a4a5df600cf72b182fd596a5004ac0d1",
- "sha256": "bc4ad7a6280091680f7a00e6d83fac1b9bbc762d87fcc339b439013978d2fe36",
- "prefix": "agent_profile_id: nu",
- "encrypted": false
- },
- "pass": true
- }
- ]
-}
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-only.gif b/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-only.gif
deleted file mode 100644
index 74d306c89a..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-only.gif and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-only.mp4 b/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-only.mp4
deleted file mode 100644
index 2f6fc36831..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-only.mp4 and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-result.png b/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-result.png
deleted file mode 100644
index cdc4dd11ca..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/encryption-root/desktop-key-result.png and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/phone-export-results.json b/.pr/map-maintenance/2026-10-10/encryption-root/phone-export-results.json
deleted file mode 100644
index f45af8faf5..0000000000
--- a/.pr/map-maintenance/2026-10-10/encryption-root/phone-export-results.json
+++ /dev/null
@@ -1,58 +0,0 @@
-{
- "viewport": "phone",
- "selectedBackend": "Local",
- "automationVersion": "1.19.3",
- "results": [
- {
- "action": "key-only set",
- "dirtyCount": 1,
- "before": {
- "sha": "dc1941b3a4a5df600cf72b182fd596a5004ac0d1",
- "sha256": "bc4ad7a6280091680f7a00e6d83fac1b9bbc762d87fcc339b439013978d2fe36",
- "prefix": "agent_profile_id: nu",
- "encrypted": false
- },
- "after": {
- "sha": "cb702b7229229c89e7da3ae45190f98c1ecffd6e",
- "sha256": "d5742257f189e4e90aa3010420f78d61bbb0e1194a6c05123c42447b99d4de62",
- "prefix": "gAAAAABqyednyKzzirJt",
- "encrypted": true
- },
- "pass": true
- },
- {
- "action": "key-only rotate",
- "dirtyCount": 1,
- "before": {
- "sha": "cb702b7229229c89e7da3ae45190f98c1ecffd6e",
- "sha256": "d5742257f189e4e90aa3010420f78d61bbb0e1194a6c05123c42447b99d4de62",
- "prefix": "gAAAAABqyednyKzzirJt",
- "encrypted": true
- },
- "after": {
- "sha": "4549dd2653504f2133a8971cc874687335d6e4be",
- "sha256": "cbbf27aa096d5f8415431dade45b212be8f00f105aaff4310155d65863ecd9cb",
- "prefix": "gAAAAABqyedsYS_9FEyJ",
- "encrypted": true
- },
- "pass": true
- },
- {
- "action": "key-only clear",
- "dirtyCount": 1,
- "before": {
- "sha": "4549dd2653504f2133a8971cc874687335d6e4be",
- "sha256": "cbbf27aa096d5f8415431dade45b212be8f00f105aaff4310155d65863ecd9cb",
- "prefix": "gAAAAABqyedsYS_9FEyJ",
- "encrypted": true
- },
- "after": {
- "sha": "1bf065bd7ee6a3a9633ce0d381ab927948010a9a",
- "sha256": "bc4ad7a6280091680f7a00e6d83fac1b9bbc762d87fcc339b439013978d2fe36",
- "prefix": "agent_profile_id: nu",
- "encrypted": false
- },
- "pass": true
- }
- ]
-}
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-only.gif b/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-only.gif
deleted file mode 100644
index b12112b2fc..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-only.gif and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-only.mp4 b/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-only.mp4
deleted file mode 100644
index 533954b9d2..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-only.mp4 and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-result.png b/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-result.png
deleted file mode 100644
index f344f7eb96..0000000000
Binary files a/.pr/map-maintenance/2026-10-10/encryption-root/phone-key-result.png and /dev/null differ
diff --git a/.pr/map-maintenance/2026-10-10/maintenance-report.md b/.pr/map-maintenance/2026-10-10/maintenance-report.md
deleted file mode 100644
index 886e787ef5..0000000000
--- a/.pr/map-maintenance/2026-10-10/maintenance-report.md
+++ /dev/null
@@ -1,63 +0,0 @@
-I'm an AI agent (Codex) helping Engel Nyst (@enyst) with project work.
-
-# Delta maintenance: verified map corrections, baseline retained
-
-BASE: 553d4519113192a80fb18376fdb6c1e39fa06387.
-TARGET: 5a8c3832154b14d2dad651801c0ab3f74c96f02e.
-
-The current PR incorporates genuine DeepSeek verification of assisted automation forms, prompt/plugin execution, task outcomes, exports, debug launch and long finish logs. The model-backed checks ran on fresh Linux stacks selecting the Local OpenHands backend, with SDK 1.54.0 and automation 1.19.3. Their 45 entry/check scopes yielded **40 pass, 4 fail, 1 blocked**; the remaining producer requirement is a genuine legacy/custom-schema finish. DeepSeek assistance and execution are verified paths in the maintained map.
-
-This is a DELTA pass. Twelve first-parent commits changed 172 paths; all resolved to merged PRs, all 89 map references were audited in their owning repositories, and source coverage includes F01–F27 (745 stable IDs). SDK/client/automation changes and shared consumers widen the required delta to all 27 families. Live coverage remains scoped, so BASE is unchanged.
-
-Covered merged PRs: [18228](https://github.com/OpenHands/OpenHands/pull/18228), [18219](https://github.com/OpenHands/OpenHands/pull/18219), [18243](https://github.com/OpenHands/OpenHands/pull/18243), [17758](https://github.com/OpenHands/OpenHands/pull/17758), [18237](https://github.com/OpenHands/OpenHands/pull/18237), [18175](https://github.com/OpenHands/OpenHands/pull/18175), [18253](https://github.com/OpenHands/OpenHands/pull/18253), [17199](https://github.com/OpenHands/OpenHands/pull/17199), [17200](https://github.com/OpenHands/OpenHands/pull/17200), [17201](https://github.com/OpenHands/OpenHands/pull/17201), [17242](https://github.com/OpenHands/OpenHands/pull/17242), [17688](https://github.com/OpenHands/OpenHands/pull/17688).
-
-## Current verified behavior and map changes
-
-- **F22.agent-assisted-setup:** the map contains literal credential-free draft save/reload/resume/custom-script tests and the exercised genuine model commands. Real `automation_form_update` filled prompt and plugin forms; desktop Hide/Show, phone Form/Conversation, associated-draft resumption, final creation and ordinary prompt/plugin Run now completed successfully. Fresh phone creation is recorded separately from resizing a desktop form. Final assisted creations are inactive until enabled; ordinary detail polling becomes Successful without reload. Draft Test's automatic completion remains a known failure with separate reload recovery.
-- **F22.setup-review-confirm:** desktop Prompt Review/Back/Confirm retained values and created the selected flash automation. Separate desktop and phone prompt creation/execution checks passed. Agent-created plugins loaded the public `magic-test` fixture and registered `magic-word`; plugin skill invocation and the distinct template Plugin review/timezone path remain separate checks.
-- **F23:** genuine successful runs updated status, summary, cost and conversation links without reload; conversation links and JSON/CSV exports worked on desktop/phone. Genuine `blocked`, `partial_success` and `unknown` finishes rendered Blocked, Partial and Needs review. Failed-run Debug launched the correctly seeded real model conversation and was paused after initial execution; no completed repair is claimed.
-- **F23.run-logs-modal:** a real finish produced a 4,596-character outcome summary and 1,079-character metadata message. Summary/metadata endings, Output/Error tabs, stdout end and close controls were reachable at 1440×1000, 390×844, 1440×500 and 390×500; the header remained visible and phone had no horizontal overflow. This verifies [#18173](https://github.com/OpenHands/OpenHands/issues/18173) / [fix #18175](https://github.com/OpenHands/OpenHands/pull/18175) on pinned Linux Local, and the map states the passing contract.
-- **F24.encryption:** key-only set/rotate/clear passes desktop/phone on pinned automation 1.19.3 in two fresh stacks. Each save dirties the export; without an intervening automation PATCH, the next sync commits ciphertext, different ciphertext, then plaintext. The maintained recipe reflects [automation#563](https://github.com/OpenHands/automation/pull/563) / [#551](https://github.com/OpenHands/automation/issues/551). [Desktop readbacks](encryption-root/desktop-export-results.json), [phone readbacks](encryption-root/phone-export-results.json), [desktop recording](encryption-root/desktop-key-only.mp4), [phone recording](encryption-root/phone-key-only.mp4). The F24 peer covered 26 IDs through 28 scopes: 25 pass, 2 Cloud prerequisites blocked and 1 non-404 status-error prerequisite not-run, including unsupported UI on automation 1.7.1.
-- **README:** verified form-tool and split-view paths are linked to F22. Remaining home handoff, full draft-management, replacement editor, ACP and Cloud boundaries are named with their actual prerequisites.
-- **F26:** seven native terminal scopes across five IDs pass (version/-v, info, help, two flag-conflict cases and missing-build scratch copy). These are CLI proof; Docker, Electron and embedded-host execution need their own checks.
-
-The credential-free F22 checks also verified draft save/reload/resume, phone draft rename/Delete, template/import/form/responder behavior and generated callback-only script execution. [Desktop draft save](draft-root/draft-desktop.mp4), [phone resume](draft-root/resume-phone.mp4) and [phone renamed draft](draft-root/resumed-phone.png) retain the genuine captures. Their viewport comes from the media, not a later ledger registration.
-
-## Scoped genuine model results
-
-| Agent assignment | Entry/check scopes | Pass | Fail | Blocked |
-|---|---:|---:|---:|---:|
-| Assisted forms | 13 | 12 | 1 | 0 |
-| Execution and outcomes | 27 | 23 | 3 | 1 |
-| Long finish logs and desktop completion polling | 5 | 5 | 0 | 0 |
-| Total | 45 | 40 | 4 | 1 |
-
-[Detailed model results](deepseek-cloud-results.md) and [filtered source receipts](deepseek-cloud-receipts.json) give the tested entries, final agent counts and ledger hashes. These are entry/check scopes, not distinct complete-ID passes. Credential-free F22, F24 and F26 checks overlap other scopes and are not added into a claimed 745-ID coverage total. No full-family or full-delta pass is inferred.
-
-## Current defects
-
-- [#18263](https://github.com/OpenHands/OpenHands/issues/18263): draft Test stays Pending after backend COMPLETED. Independently verified with callback-only scripts, a real desktop prompt test and a real phone plugin test. The desktop model test stayed Pending through a 45.263-second Successful-row wait; the phone capture was 201.362 seconds after completion. Reload recovered Successful. The map preserves automatic completion as Expected and documents the failure. [First script backend receipt](../../bugs/2026-10-10/draft-run-status/first/sanitized-run-summary.json) and [second receipt](../../bugs/2026-10-10/draft-run-status/second/second-sanitized-run-summary.json) remain accessible.
-- [#17948](https://github.com/OpenHands/OpenHands/issues/17948): a real timeout-1 prompt run ended FAILED without a conversation, but the desktop/phone activity row called it a script. This result covers the labeling symptom.
-- F22's existing [#17959](https://github.com/OpenHands/OpenHands/issues/17959) blank no-match search and [#18061](https://github.com/OpenHands/OpenHands/issues/18061) blank Default after kind switches remain failures.
-
-Existing issues are linked; no duplicates or product fixes are included. No before-change live comparison establishes a causal regression.
-
-## Remaining prerequisites and coverage
-
-The model-result table's single block needs a **genuine legacy/custom-schema finish producer** for absent/non-string/unrecognized status fallback, plus an actual failed producer carrying a finish summary for failed-run summary retention. The normal TaskOutcome producer proves its supported enum statuses. Injected run metadata or mocked model output is not live proof.
-
-Other remaining coverage includes the home Code/Automate handoff, complete dashboard draft management and replacement editor; template Plugin review/timezone and plugin skill invocation; Pi/OpenCode ACP presets and genuine execution; Cloud organization handback, enterprise setup-state/tour entitlements, member permissions and multi-organization Git Sync; native/Docker/embedded paths and the rest of the affected recipes. F24.error-state needs a genuine non-404 status failure while automation remains healthy. The map index records these boundaries.
-
-Unmapped-path accounting: scripts/check-sdk-version-sync.mjs is a contributor gate; acp-brand-marks.ts belongs to ACP icon consumers; automation-form.ts is shared by composer/conversation and automation-form consumers. F22 names the form constant/session/service and exercises the form tool and split navigation.
-
-## Runtime, validation and teardown
-
-Canvas 1.26.0 product source was built at TARGET. Genuine model verification used documentation head `ac3ea58e42d09178f027878a66b7b2e36bf59abd`, whose product source matches TARGET, with Node 24.19.0/npm 11.9.0, Chromium 151.0.7922.173, SDK 1.54.0 and automation 1.19.3; build identity `31b4983403910d82`. Credential-free and Git Sync captures used macOS 27.0.1/Chrome 155.0.8059.39, Node 25.9.0/npm 11.12.1 and uv 0.11.19, with the same backend pins. Builds and doctor passed. Default themes/fonts follow each recorded run; font metrics were not separately enumerated.
-
-Reported model SDK estimates are listed in the detailed results; validation/title/helper costs are not fully accounted, so this is not a test of the scheduler's aggregate budget enforcement. The supplied DeepSeek key and real model calls succeeded.
-
-Map structure, route/component coverage, cited test IDs, BASE ancestry and explicit BASE/TARGET affected checks pass. These static gates validate the map and are separate from live behavior.
-
-Owned automations/drafts were deleted, Git Sync disabled/cleared, and retained evidence survived teardown. Linux model stacks had no live owned-group processes and all ports refused connections; unreaped dead zombie PIDs account for CLI `launcherStopped=false`. The cloud root reviewed the original private captures; published model receipts are filtered transcriptions of its completed reports/outputs. Credential values, raw system prompts, browser state and private homes remain unpublished. The source session scanned 384 evidence text files with no matches against its two configured credential values.
-
-The accepted BASE remains unchanged because the full delta and its distinct backend/platform entry points are incomplete.