Skip to content

YC (S26) | Open Computer History | Continuously record your company computer work, map your workflows, help you find work worth automating, and power your agents' context

fix: preserve ACP tool metadata in chat

Louis Beaumont7b0e80820c02
diff --git a/apps/screenpipe-app-tauri/components/chat/standalone/hooks/use-pi-foreground-events.ts b/apps/screenpipe-app-tauri/components/chat/standalone/hooks/use-pi-foreground-events.ts
index e8c1766c1..49b0930e4 100644
--- a/apps/screenpipe-app-tauri/components/chat/standalone/hooks/use-pi-foreground-events.ts
+++ b/apps/screenpipe-app-tauri/components/chat/standalone/hooks/use-pi-foreground-events.ts
@@ -491,6 +491,7 @@ export function usePiForegroundEvents({
if (piMessageIdRef.current) {
const msgId = piMessageIdRef.current;
const parentToolCallId = stringValue(data.parentToolCallId);
+ const agentId = stringValue(data.agentId);
const toolKind = stringValue(data.kind);
const toolCall: ToolCall = {
id: stringValue(data.toolCallId, Date.now().toString()),
@@ -498,6 +499,7 @@ export function usePiForegroundEvents({
args: isRecord(data.args) ? data.args : {},
isRunning: true,
startedAtMs: Date.now(),
+ ...(agentId ? { agentId } : {}),
...(toolKind ? { kind: toolKind } : {}),
...(parentToolCallId ? { parentToolCallId } : {}),
...(data.subagent === true ? { subagent: true } : {}),
@@ -543,6 +545,11 @@ export function usePiForegroundEvents({
if (data.retry !== undefined) block.toolCall.retry = data.retry;
const title = stringValue(data.title);
if (title) block.toolCall.toolName = title;
+ const agentId = stringValue(data.agentId);
+ if (agentId) block.toolCall.agentId = agentId;
+ const kind = stringValue(data.kind);
+ if (kind) block.toolCall.kind = kind;
+ if (isRecord(data.args)) block.toolCall.args = data.args;
const outputDelta = stringValue(data.outputDelta);
if (outputDelta) {
const combined = `${block.toolCall.progress ?? ""}${outputDelta}`;
@@ -566,6 +573,13 @@ export function usePiForegroundEvents({
// Update the tool block in-place
for (const block of piContentBlocksRef.current) {
if (block.type === "tool" && block.toolCall.id === toolCallId) {
+ const toolName = stringValue(data.toolName);
+ if (toolName) block.toolCall.toolName = toolName;
+ const agentId = stringValue(data.agentId);
+ if (agentId) block.toolCall.agentId = agentId;
+ const kind = stringValue(data.kind);
+ if (kind) block.toolCall.kind = kind;
+ if (isRecord(data.args)) block.toolCall.args = data.args;
block.toolCall.isRunning = false;
block.toolCall.result = truncated;
block.toolCall.isError = data.isError === true;
diff --git a/apps/screenpipe-app-tauri/components/chat/standalone/message-content.tsx b/apps/screenpipe-app-tauri/components/chat/standalone/message-content.tsx
index 926cc3204..9df19b887 100644
--- a/apps/screenpipe-app-tauri/components/chat/standalone/message-content.tsx
+++ b/apps/screenpipe-app-tauri/components/chat/standalone/message-content.tsx
@@ -1368,11 +1368,16 @@ function ToolActivityGroup({
// Tool activity is progressively disclosed: one friendly status is visible
// by default, and the user can expand a high-level activity list. Interactive
// tools stay open because hiding a question would block the conversation.
+ const e2eForceExpanded =
+ process.env.NEXT_PUBLIC_SCREENPIPE_E2E === "true" &&
+ typeof window !== "undefined" &&
+ (window as typeof window & { __e2eExpandToolActivity?: boolean })
+ .__e2eExpandToolActivity === true;
const isExpanded = forceCollapsed
? false
: hideSummary
? true
- : hasInteractiveTool
+ : hasInteractiveTool || e2eForceExpanded
? true
: manualExpand !== null ? manualExpand : defaultExpanded;
@@ -1438,10 +1443,10 @@ function ToolActivityGroup({
<AnimatePresence>
{isExpanded && (
<motion.div
- initial={{ height: 0, opacity: 0 }}
+ initial={e2eForceExpanded ? false : { height: 0, opacity: 0 }}
animate={{ height: "auto", opacity: 1 }}
exit={{ height: 0, opacity: 0 }}
- transition={{ duration: reduceMotion ? 0 : 0.15 }}
+ transition={{ duration: reduceMotion || e2eForceExpanded ? 0 : 0.15 }}
className="overflow-hidden"
>
<div
@@ -1484,9 +1489,12 @@ function ToolActivityGroup({
return (
<motion.div
key={toolCallRenderKey(tc, i)}
- initial={reduceMotion ? false : { opacity: 0, x: -8 }}
+ initial={reduceMotion || e2eForceExpanded ? false : { opacity: 0, x: -8 }}
animate={{ opacity: 1, x: 0 }}
- transition={{ duration: reduceMotion ? 0 : 0.15, delay: reduceMotion ? 0 : i * 0.03 }}
+ transition={{
+ duration: reduceMotion || e2eForceExpanded ? 0 : 0.15,
+ delay: reduceMotion || e2eForceExpanded ? 0 : i * 0.03,
+ }}
>
<ToolCallRailItem
toolCall={tc}
diff --git a/apps/screenpipe-app-tauri/e2e/COVERAGE.md b/apps/screenpipe-app-tauri/e2e/COVERAGE.md
index 3da4eb16b..7f2e12fe1 100644
--- a/apps/screenpipe-app-tauri/e2e/COVERAGE.md
+++ b/apps/screenpipe-app-tauri/e2e/COVERAGE.md
@@ -7,8 +7,8 @@ and layer declared in the manifest, weighted by confidence and criticality.
- Manifest: `e2e/coverage-map.json`
- Specs directory: `e2e/specs`
- Mapped specs: 120
-- Declared test blocks: 351
-- Weighted coverage points: 275.5
+- Declared test blocks: 352
+- Weighted coverage points: 276.2
Confidence weights: strong=1.0, partial=0.7, conditional=0.4, smoke=0.3.
Criticality weights: high=1.0, medium=0.7, low=0.4.
@@ -19,9 +19,9 @@ can execute more runtime cases than this number shows.
| Platform | Specs | Declared tests | Weighted points | Layers | Features | Critical score |
| --- | --- | --- | --- | --- | --- | --- |
-| windows | 91 | 301 | 245.9 | 15 | 98 | 92% |
-| macos | 116 | 313 | 245.3 | 17 | 105 | 90% |
-| linux | 80 | 259 | 215.3 | 14 | 95 | 88% |
+| windows | 91 | 302 | 246.6 | 15 | 99 | 92% |
+| macos | 116 | 314 | 246.0 | 17 | 106 | 90% |
+| linux | 80 | 260 | 216.0 | 14 | 96 | 88% |
## Runtime Results
@@ -37,18 +37,18 @@ pass/fail/skip counts.
| auth | - | 1 specs / 1 tests / 1.0 pts | - |
| billing | 4 specs / 6 tests / 5.7 pts | 4 specs / 6 tests / 5.7 pts | 4 specs / 6 tests / 5.7 pts |
| capture-ocr | 2 specs / 16 tests / 6.4 pts | 8 specs / 12 tests / 4.8 pts | 1 specs / 3 tests / 1.2 pts |
-| chat-ai | 26 specs / 57 tests / 43.1 pts | 38 specs / 79 tests / 57.2 pts | 25 specs / 56 tests / 42.6 pts |
+| chat-ai | 26 specs / 58 tests / 43.8 pts | 38 specs / 80 tests / 57.9 pts | 25 specs / 57 tests / 43.3 pts |
| entitlement | - | 1 specs / 1 tests / 1.0 pts | - |
| local-api | 25 specs / 114 tests / 95.0 pts | 33 specs / 104 tests / 87.5 pts | 20 specs / 82 tests / 73.2 pts |
| notifications | 4 specs / 26 tests / 17.3 pts | 3 specs / 5 tests / 3.4 pts | 2 specs / 4 tests / 3.1 pts |
| onboarding | 9 specs / 37 tests / 32.8 pts | 11 specs / 39 tests / 34.2 pts | 9 specs / 37 tests / 32.8 pts |
-| os-integration | 7 specs / 30 tests / 25.5 pts | 14 specs / 27 tests / 16.0 pts | 2 specs / 13 tests / 9.4 pts |
+| os-integration | 7 specs / 31 tests / 26.2 pts | 14 specs / 28 tests / 16.7 pts | 2 specs / 14 tests / 10.1 pts |
| performance | 2 specs / 44 tests / 44.0 pts | 4 specs / 34 tests / 30.5 pts | 1 specs / 29 tests / 29.0 pts |
| pipes | 6 specs / 19 tests / 19.0 pts | 8 specs / 25 tests / 25.0 pts | 6 specs / 19 tests / 19.0 pts |
| real-ui-e2e | 65 specs / 202 tests / 166.5 pts | 79 specs / 210 tests / 172.4 pts | 60 specs / 178 tests / 152.6 pts |
| settings | 14 specs / 40 tests / 37.0 pts | 16 specs / 34 tests / 29.7 pts | 13 specs / 31 tests / 28.0 pts |
| storage-privacy | 9 specs / 42 tests / 33.3 pts | 9 specs / 27 tests / 26.1 pts | 6 specs / 20 tests / 19.1 pts |
-| tauri-command | 19 specs / 52 tests / 40.1 pts | 27 specs / 64 tests / 48.3 pts | 18 specs / 53 tests / 40.9 pts |
+| tauri-command | 19 specs / 53 tests / 40.8 pts | 27 specs / 65 tests / 49.0 pts | 18 specs / 54 tests / 41.6 pts |
| window-lifecycle | 19 specs / 66 tests / 55.0 pts | 19 specs / 46 tests / 32.4 pts | 13 specs / 39 tests / 29.9 pts |
## Critical Feature Matrix
@@ -100,7 +100,7 @@ pass/fail/skip counts.
| Spec | Platforms | Layers | Features | Criticality | Confidence | UX | Tests | Notes |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
-| acp-backend.spec.ts | windows, macos, linux | chat-ai, tauri-command, os-integration | chat, acp-backend, agent-streaming, agent-permissions, agent-cancellation, agent-terminal, mcp-registration, process-supervision | high | partial | mixed | 12 | Official Rust ACP SDK runtime and inline UI E2E with a deterministic stdio agent: initialization, session close, streaming, background permissions, cancellation, localized auth/cancel, malformed stdout, exact adapter/descendant reaping, terminal lifecycle/output draining, and a real authenticated local API MCP health call. Curated third-party adapters remain manual smoke coverage. |
+| acp-backend.spec.ts | windows, macos, linux | chat-ai, tauri-command, os-integration | chat, acp-backend, agent-streaming, agent-permissions, agent-cancellation, agent-terminal, agent-tool-metadata, mcp-registration, process-supervision | high | partial | mixed | 13 | Official Rust ACP SDK runtime and inline UI E2E with a deterministic stdio agent: initialization, session close, streaming, background permissions, cancellation, localized auth/cancel, malformed stdout, exact adapter/descendant reaping, terminal lifecycle/output draining, late tool metadata/error propagation with a provider fallback, and a real authenticated local API MCP health call. Curated third-party adapters remain manual smoke coverage. |
| acp-onboarding-ux.spec.ts | windows, macos, linux | chat-ai, real-ui-e2e, tauri-command | chat, acp-backend, acp-onboarding, acp-auth | high | partial | real-user-flow | 2 | Selecting an ACP agent connects it with nothing sent (sign-in surfaces before the first send, not after a cold install), and the settings picker states who owns sign-in/models/API keys next to the agent list. |
| api-key-cold-spawn.spec.ts | windows, macos, linux | local-api, tauri-command | local-api-auth, app-launch | medium | partial | command | 3 | Cold-spawn local API config regression coverage. |
| api-search-stress.spec.ts | windows, macos, linux | local-api, performance | local-api-auth, local-api-search, health, audio-device-health, local-api-load | high | strong | api | 29 | Broad readonly API, auth, search, and load coverage. |
diff --git a/apps/screenpipe-app-tauri/e2e/coverage-map.json b/apps/screenpipe-app-tauri/e2e/coverage-map.json
index 43e64f587..9ac47def1 100644
--- a/apps/screenpipe-app-tauri/e2e/coverage-map.json
+++ b/apps/screenpipe-app-tauri/e2e/coverage-map.json
@@ -336,13 +336,14 @@
"agent-permissions",
"agent-cancellation",
"agent-terminal",
+ "agent-tool-metadata",
"mcp-registration",
"process-supervision"
],
"criticality": "high",
"confidence": "partial",
"ux": "mixed",
- "notes": "Official Rust ACP SDK runtime and inline UI E2E with a deterministic stdio agent: initialization, session close, streaming, background permissions, cancellation, localized auth/cancel, malformed stdout, exact adapter/descendant reaping, terminal lifecycle/output draining, and a real authenticated local API MCP health call. Curated third-party adapters remain manual smoke coverage."
+ "notes": "Official Rust ACP SDK runtime and inline UI E2E with a deterministic stdio agent: initialization, session close, streaming, background permissions, cancellation, localized auth/cancel, malformed stdout, exact adapter/descendant reaping, terminal lifecycle/output draining, late tool metadata/error propagation with a provider fallback, and a real authenticated local API MCP health call. Curated third-party adapters remain manual smoke coverage."
},
{
"spec": "acp-onboarding-ux.spec.ts",
diff --git a/apps/screenpipe-app-tauri/e2e/fixtures/mock-acp-agent.ts b/apps/screenpipe-app-tauri/e2e/fixtures/mock-acp-agent.ts
index e78fdf4dd..7d63620b6 100644
--- a/apps/screenpipe-app-tauri/e2e/fixtures/mock-acp-agent.ts
+++ b/apps/screenpipe-app-tauri/e2e/fixtures/mock-acp-agent.ts
@@ -17,7 +17,7 @@ type JsonRpcMessage = {
error?: unknown;
};
-type Scenario = "normal" | "malformed" | "exit" | "auth" | "mcp" | "tree" | "terminal" | "subagent" | "resume";
+type Scenario = "normal" | "malformed" | "exit" | "auth" | "mcp" | "tree" | "terminal" | "subagent" | "metadata" | "resume";
const scenarioArg = process.argv.find((arg) => arg.startsWith("--scenario="));
const scenario = (scenarioArg?.slice("--scenario=".length) ?? "normal") as Scenario;
@@ -306,6 +306,45 @@ function emitPromptPrelude(): void {
});
}
+function emitMetadataRefinementFlow(requestId: JsonRpcId): void {
+ update({
+ sessionUpdate: "tool_call",
+ toolCallId: "mock-refined",
+ title: "MCP: tool",
+ status: "in_progress",
+ rawInput: {},
+ });
+ update({
+ sessionUpdate: "tool_call_update",
+ toolCallId: "mock-refined",
+ title: "mcp__screenpipe__search-content",
+ kind: "search",
+ status: "completed",
+ rawInput: { query: "late ACP metadata" },
+ rawOutput: { error: "Tool execution error" },
+ });
+ update({
+ sessionUpdate: "tool_call",
+ toolCallId: "mock-provider-fallback",
+ title: "MCP: tool",
+ status: "in_progress",
+ rawInput: {},
+ });
+ update({
+ sessionUpdate: "tool_call_update",
+ toolCallId: "mock-provider-fallback",
+ status: "completed",
+ rawOutput: { success: true },
+ });
+ update({
+ sessionUpdate: "agent_message_chunk",
+ content: { type: "text", text: "Metadata refinement turn complete." },
+ messageId: "mock-metadata-final",
+ });
+ respond(requestId, { stopReason: "end_turn" });
+ activePromptRequestId = undefined;
+}
+
/** Deterministic subagent turn matching the claude-agent-acp and codex-acp
* wire formats: a parent Task tool call, a flat child stamped with
* _meta.claudeCode.parentToolUseId, heartbeat and streamed-output
@@ -612,6 +651,10 @@ async function handleRequest(message: JsonRpcMessage): Promise<void> {
emitSubagentFlow(message.id);
return;
}
+ if (scenario === "metadata") {
+ emitMetadataRefinementFlow(message.id);
+ return;
+ }
if (!activePromptIsCancellation) beginPermissionFlow();
return;
}
diff --git a/apps/screenpipe-app-tauri/e2e/specs/acp-backend.spec.ts b/apps/screenpipe-app-tauri/e2e/specs/acp-backend.spec.ts
index 45b3e8dca..4e8093712 100644
--- a/apps/screenpipe-app-tauri/e2e/specs/acp-backend.spec.ts
+++ b/apps/screenpipe-app-tauri/e2e/specs/acp-backend.spec.ts
@@ -10,6 +10,7 @@ import { fileURLToPath } from "node:url";
import { openHomeWindow, waitForAppReady, t } from "../helpers/test-utils.js";
import { invokeOrThrow } from "../helpers/tauri.js";
+import { saveScreenshot } from "../helpers/screenshot-utils.js";
type AgentEnvelope = {
source?: string;
@@ -42,7 +43,7 @@ type ProcessMarker = {
descendantPid?: number;
};
-type AcpScenario = "normal" | "malformed" | "exit" | "auth" | "mcp" | "tree" | "terminal" | "subagent" | "resume";
+type AcpScenario = "normal" | "malformed" | "exit" | "auth" | "mcp" | "tree" | "terminal" | "subagent" | "metadata" | "resume";
const fixturePath = fileURLToPath(new URL("../fixtures/mock-acp-agent.ts", import.meta.url));
let normalSession = "";
@@ -55,6 +56,7 @@ let treeSession = "";
let mcpSession = "";
let terminalSession = "";
let subagentSession = "";
+let metadataSession = "";
let resumeSession = "";
let orphanedPresetSession = "";
let treeMarkerPrefix = "";
@@ -71,6 +73,7 @@ function resetRunIdentifiers(): void {
mcpSession = randomUUID();
terminalSession = randomUUID();
subagentSession = randomUUID();
+ metadataSession = randomUUID();
resumeSession = randomUUID();
orphanedPresetSession = randomUUID();
treeMarkerPrefix = path.join(os.tmpdir(), `screenpipe-acp-process-${treeSession}`);
@@ -85,7 +88,7 @@ function acpProviderConfig(
return {
backend: "acp",
acpAgent: {
- id: "custom",
+ id: scenario === "metadata" ? "pi-acp" : "custom",
command: process.execPath,
args: [fixturePath, `--scenario=${scenario}`],
env,
@@ -304,7 +307,7 @@ async function answerAgentAction(
async function startAcp(
sessionId: string,
- scenario: "normal" | "malformed" | "mcp" | "tree" | "terminal" | "subagent" | "resume",
+ scenario: "normal" | "malformed" | "mcp" | "tree" | "terminal" | "subagent" | "metadata" | "resume",
env: Record<string, string> = {},
userToken: string | null = null,
resumeSessionId: string | null = null,
@@ -552,6 +555,7 @@ describe("ACP backend", function () {
await invokeOrThrow("pi_stop", { sessionId: treeSession }).catch(() => undefined);
await invokeOrThrow("pi_stop", { sessionId: mcpSession }).catch(() => undefined);
await invokeOrThrow("pi_stop", { sessionId: terminalSession }).catch(() => undefined);
+ await invokeOrThrow("pi_stop", { sessionId: metadataSession }).catch(() => undefined);
rmSync(processMarkerPath("adapter"), { force: true });
rmSync(processMarkerPath("descendant"), { force: true });
});
@@ -913,6 +917,63 @@ describe("ACP backend", function () {
await stopAndAssertGone(subagentSession);
});
+ it("preserves terminal ACP metadata and renders traceable tool activity", async () => {
+ await startAcp(metadataSession, "metadata");
+ await foregroundChat(metadataSession);
+ await browser.execute(() => {
+ (window as any).__e2eExpandToolActivity = true;
+ });
+ await beginPrompt(metadataSession, "refine late tool metadata");
+ const prompt = await waitForPromptDone();
+ expect(prompt.error).toBeUndefined();
+ await browser.waitUntil(
+ async () => (await capturedEvents(metadataSession)).some(
+ (envelope) => envelope.event?.type === "agent_end",
+ ),
+ {
+ timeout: t(10_000),
+ interval: 100,
+ timeoutMsg: "metadata refinement turn did not complete",
+ },
+ );
+
+ const widgets = await $$('[data-testid="tool-activity-widget"]');
+ expect(widgets.length).toBeGreaterThan(0);
+ const widget = widgets[widgets.length - 1];
+ const list = await widget.$('[data-testid="tool-activity-list"]');
+ await list.waitForExist({ timeout: t(10_000) });
+ const rows = await widget.$$('[data-testid="tool-activity-item"]');
+ expect(rows).toHaveLength(2);
+ const listText = await list.getText();
+ expect(listText).toContain("Searched your history");
+ expect(listText).toContain("Used a Pi tool");
+ expect(listText).not.toContain("Completed a background step");
+
+ await rows[0].click();
+ await browser.waitUntil(async () => (await widget.getText()).includes("late ACP metadata"), {
+ timeout: t(5_000),
+ interval: 100,
+ timeoutMsg: "refined ACP arguments were not disclosed in the tool row",
+ });
+ const variant = process.env.SCREENPIPE_PR_PROOF_VARIANT ?? "after";
+ const screenshot = await saveScreenshot(`acp-tool-metadata-${variant}`);
+ expect(existsSync(screenshot)).toBe(true);
+
+ const terminal = (await capturedEvents(metadataSession)).find(
+ (envelope) =>
+ envelope.event?.type === "tool_execution_end" &&
+ envelope.event?.toolCallId === "mock-refined",
+ )?.event;
+ expect(terminal).toMatchObject({
+ toolName: "mcp__screenpipe__search-content",
+ agentId: "pi-acp",
+ kind: "search",
+ args: { query: "late ACP metadata" },
+ isError: true,
+ });
+ await stopAndAssertGone(metadataSession);
+ });
+
it("reattaches to a prior session via session/resume when one is provided", async () => {
await startAcp(resumeSession, "resume", {}, null, "mock-acp-session");
const events = await capturedEvents(resumeSession);
diff --git a/apps/screenpipe-app-tauri/lib/__tests__/pi-event-router.test.ts b/apps/screenpipe-app-tauri/lib/__tests__/pi-event-router.test.ts
index fea0b0457..9590804db 100644
--- a/apps/screenpipe-app-tauri/lib/__tests__/pi-event-router.test.ts
+++ b/apps/screenpipe-app-tauri/lib/__tests__/pi-event-router.test.ts
@@ -953,6 +953,48 @@ describe("pi-event-router: subagent tool progress", () => {
).toolCall;
expect(cmd.progress.length).toBe(4000);
});
+
+ it("keeps late ACP metadata on a completed background tool", async () => {
+ seed("A");
+ useChatStore.setState({ currentId: "B" });
+ await handlePiEvent(
+ piEvt("A", { type: "message_start", message: { role: "assistant" } }),
+ );
+ await handlePiEvent(
+ piEvt("A", {
+ type: "tool_execution_start",
+ toolCallId: "late-metadata",
+ toolName: "MCP: tool",
+ agentId: "cursor",
+ args: {},
+ } as unknown as AgentInnerEvent),
+ );
+ await handlePiEvent(
+ piEvt("A", {
+ type: "tool_execution_end",
+ toolCallId: "late-metadata",
+ toolName: "mcp__screenpipe__search-content",
+ agentId: "cursor",
+ kind: "search",
+ args: { query: "late ACP metadata" },
+ result: { content: [{ type: "text", text: '{"error":"Tool execution error"}' }] },
+ isError: true,
+ } as unknown as AgentInnerEvent),
+ );
+
+ const session = useChatStore.getState().sessions.A;
+ const tool = (session.contentBlocks as any[]).find(
+ (block) => block.type === "tool" && block.toolCall.id === "late-metadata",
+ ).toolCall;
+ expect(tool).toMatchObject({
+ toolName: "mcp__screenpipe__search-content",
+ agentId: "cursor",
+ kind: "search",
+ args: { query: "late ACP metadata" },
+ isRunning: false,
+ isError: true,
+ });
+ });
});
describe("pi-event-router: honest interruption on quit/crash", () => {
diff --git a/apps/screenpipe-app-tauri/lib/chat/__tests__/tool-presentation.test.ts b/apps/screenpipe-app-tauri/lib/chat/__tests__/tool-presentation.test.ts
index dbe466ea2..8a65ea897 100644
--- a/apps/screenpipe-app-tauri/lib/chat/__tests__/tool-presentation.test.ts
+++ b/apps/screenpipe-app-tauri/lib/chat/__tests__/tool-presentation.test.ts
@@ -469,6 +469,21 @@ describe("presentToolActivity — ACP tool calls", () => {
"Completed a background step",
);
});
+
+ it("identifies the provider when ACP supplies no meaningful tool metadata", () => {
+ expect(
+ presentToolActivity({ toolName: "MCP: tool", agentId: "cursor", args: {} }),
+ ).toMatchObject({
+ runningLabel: "Using a Cursor tool",
+ completedLabel: "Used a Cursor tool",
+ });
+ expect(
+ presentToolActivity({ toolName: "MCP: tool", agentId: "pi-acp", args: {} }).completedLabel,
+ ).toBe("Used a Pi tool");
+ expect(
+ presentToolActivity({ toolName: "MCP: tool", agentId: "custom", args: {} }).completedLabel,
+ ).toBe("Used an ACP tool");
+ });
});
describe("MCP server startup diagnostics", () => {
diff --git a/apps/screenpipe-app-tauri/lib/chat/tool-presentation.ts b/apps/screenpipe-app-tauri/lib/chat/tool-presentation.ts
index 5d9b4ff11..7e4fee519 100644
--- a/apps/screenpipe-app-tauri/lib/chat/tool-presentation.ts
+++ b/apps/screenpipe-app-tauri/lib/chat/tool-presentation.ts
@@ -469,6 +469,7 @@ export interface PresentableToolCall {
kind?: string;
isRunning?: boolean;
isError?: boolean;
+ agentId?: string;
// Set when this call is a subagent launch (Claude's Task/Agent tool). It is a
// container for the nested transcript, so it is labeled as a subagent run
// rather than by its `think` kind, which reads as "Thought it through".
@@ -509,6 +510,19 @@ const GENERIC_ACTIVITY: ToolActivityPresentation = {
icon: "work",
};
+function providerToolActivity(agentId?: string): ToolActivityPresentation | null {
+ if (!agentId) return null;
+ const provider = ({
+ "cursor": "Cursor",
+ "claude-acp": "Claude",
+ "codex-acp": "Codex",
+ "pi-acp": "Pi",
+ "github-copilot-cli": "GitHub Copilot",
+ } as Record<string, string>)[agentId.toLowerCase()] ?? "ACP";
+ const article = provider === "ACP" ? "an" : "a";
+ return activity(`Using ${article} ${provider} tool`, `Used ${article} ${provider} tool`, "work");
+}
+
function activity(
runningLabel: string,
completedLabel: string,
@@ -1038,7 +1052,7 @@ export function presentToolActivity(toolCall: PresentableToolCall): ToolActivity
return activity(humanized, humanized, "work");
}
}
- return GENERIC_ACTIVITY;
+ return providerToolActivity(toolCall.agentId) ?? GENERIC_ACTIVITY;
}
/**
diff --git a/apps/screenpipe-app-tauri/lib/chat/types.ts b/apps/screenpipe-app-tauri/lib/chat/types.ts
index 566d4ee8e..f77028f79 100644
--- a/apps/screenpipe-app-tauri/lib/chat/types.ts
+++ b/apps/screenpipe-app-tauri/lib/chat/types.ts
@@ -59,6 +59,9 @@ export type ChatAttachment = {
export interface ToolCall {
id: string;
toolName: string;
+ /** ACP adapter id that produced this call. Used only for a safe provider
+ * fallback when the adapter never supplies a meaningful tool title. */
+ agentId?: string;
/** ACP tool-call kind (read/edit/execute/fetch/search/think/…) when the agent
* provides one; drives a sensible activity label for native ACP tools. */
kind?: string;
diff --git a/apps/screenpipe-app-tauri/lib/stores/pi-event-router.ts b/apps/screenpipe-app-tauri/lib/stores/pi-event-router.ts
index 81d2ce708..8f1e7a037 100644
--- a/apps/screenpipe-app-tauri/lib/stores/pi-event-router.ts
+++ b/apps/screenpipe-app-tauri/lib/stores/pi-event-router.ts
@@ -756,15 +756,20 @@ function applyEventToSessionContent(sid: string, payload: PiInnerEvent) {
if (!cur?.streamingMessageId) return;
const msgId = cur.streamingMessageId;
const parentToolCallId = (payload as any).parentToolCallId;
+ const agentId = (payload as any).agentId;
+ const toolKind = (payload as any).kind;
const tool = {
id: (payload as any).toolCallId || `${Date.now()}`,
toolName: (payload as any).toolName || "unknown",
args: (payload as any).args || {},
isRunning: true,
startedAtMs: Date.now(),
+ ...(typeof agentId === "string" && agentId ? { agentId } : {}),
+ ...(typeof toolKind === "string" && toolKind ? { kind: toolKind } : {}),
...(typeof parentToolCallId === "string" && parentToolCallId
? { parentToolCallId }
: {}),
+ ...((payload as any).subagent === true ? { subagent: true } : {}),
};
const blocks = [...((cur.contentBlocks as any[]) ?? []), { type: "tool", toolCall: tool }];
store.actions.setStreaming(sid, { contentBlocks: blocks });
@@ -817,6 +822,9 @@ function applyEventToSessionContent(sid: string, payload: PiInnerEvent) {
if (typeof p.subagentType === "string") next.subagentType = p.subagentType;
if (p.retry !== undefined) next.retry = p.retry;
if (typeof p.title === "string" && p.title) next.toolName = p.title;
+ if (typeof p.agentId === "string" && p.agentId) next.agentId = p.agentId;
+ if (typeof p.kind === "string" && p.kind) next.kind = p.kind;
+ if (p.args && typeof p.args === "object" && !Array.isArray(p.args)) next.args = p.args;
if (typeof p.outputDelta === "string" && p.outputDelta) {
const combined = `${next.progress ?? ""}${p.outputDelta}`;
next.progress = combined.length > 4000 ? combined.slice(-4000) : combined;
@@ -850,6 +858,12 @@ function applyEventToSessionContent(sid: string, payload: PiInnerEvent) {
...b,
toolCall: {
...b.toolCall,
+ ...((payload as any).toolName ? { toolName: (payload as any).toolName } : {}),
+ ...((payload as any).agentId ? { agentId: (payload as any).agentId } : {}),
+ ...((payload as any).kind ? { kind: (payload as any).kind } : {}),
+ ...((payload as any).args && typeof (payload as any).args === "object"
+ ? { args: (payload as any).args }
+ : {}),
isRunning: false,
result: truncated,
isError: (payload as any).isError,
diff --git a/crates/screenpipe-core/src/agents/acp/runtime.rs b/crates/screenpipe-core/src/agents/acp/runtime.rs
index f5fafcc32..e8170cb90 100644
--- a/crates/screenpipe-core/src/agents/acp/runtime.rs
+++ b/crates/screenpipe-core/src/agents/acp/runtime.rs
@@ -1783,7 +1783,10 @@ impl RuntimeState {
self.output.send(json!({
"type": "tool_execution_end",
"toolCallId": tool_call_id,
+ "agentId": self.agent_id,
"toolName": tool_name(&tool),
+ "kind": tool_kind(&tool),
+ "args": tool_args(&tool),
"result": result,
"isError": is_error
}));
@@ -1917,6 +1920,7 @@ impl RuntimeState {
let mut start = json!({
"type": "tool_execution_start",
"toolCallId": id,
+ "agentId": self.agent_id,
"toolName": tool_name(&update),
// The ACP category (read/edit/execute/fetch/search/...) lets
// the desktop label native agent tools whose human title
@@ -1936,7 +1940,7 @@ impl RuntimeState {
self.output.send(start);
if update_status_finished(&update) {
self.observe_provider_schedule(&update);
- finish_tool(&self.output, &id, &update);
+ finish_tool(&self.output, &self.agent_id, &id, &update);
turn.active_tools.remove(&id);
}
}
@@ -1963,6 +1967,7 @@ impl RuntimeState {
let mut start = json!({
"type": "tool_execution_start",
"toolCallId": id,
+ "agentId": self.agent_id,
"toolName": tool_name(&merged),
"kind": tool_kind(&merged),
"args": tool_args(&merged),
@@ -1975,12 +1980,13 @@ impl RuntimeState {
}
if update_status_finished(&merged) {
self.observe_provider_schedule(&merged);
- finish_tool(&self.output, &id, &merged);
+ finish_tool(&self.output, &self.agent_id, &id, &merged);
turn.active_tools.remove(&id);
} else if let Some(progress) = tool_progress(&update) {
let mut event = json!({
"type": "tool_execution_progress",
"toolCallId": id,
+ "agentId": self.agent_id,
});
// Carry the parent so a subagent heartbeat nests under its
// Task row even when linkage only appears on updates.
@@ -2308,6 +2314,12 @@ fn tool_progress(update: &Value) -> Option<Value> {
{
fields.insert("title".into(), json!(title));
}
+ if let Some(kind) = tool_kind(update) {
+ fields.insert("kind".into(), json!(kind));
+ }
+ if let Some(args) = tool_args_if_present(update) {
+ fields.insert("args".into(), args);
+ }
if fields.is_empty() {
None
} else {
@@ -2391,12 +2403,16 @@ fn tool_kind(update: &Value) -> Option<String> {
// adapters omit it and carry the invocation under `input`/`arguments`. Falling
// back keeps the desktop tool card (and its citations) from rendering empty.
fn tool_args(update: &Value) -> Value {
+ tool_args_if_present(update).unwrap_or_else(|| json!({}))
+}
+
+fn tool_args_if_present(update: &Value) -> Option<Value> {
for key in ["rawInput", "input", "arguments"] {
if let Some(object) = update.get(key).filter(|value| value.is_object()) {
- return object.clone();
+ return Some(object.clone());
}
}
- json!({})
+ None
}
fn update_status_finished(update: &Value) -> bool {
@@ -2415,8 +2431,44 @@ fn tool_result_text(update: &Value) -> String {
.unwrap_or_default()
}
-fn finish_tool(output: &ParentOutput, id: &str, update: &Value) {
- let is_error = update.get("status").and_then(Value::as_str) == Some("failed");
+fn tool_result_is_error(update: &Value) -> bool {
+ if update.get("status").and_then(Value::as_str) == Some("failed") {
+ return true;
+ }
+
+ fn explicit_error(value: &Value) -> bool {
+ let parsed;
+ let value = if let Some(text) = value.as_str() {
+ parsed = match serde_json::from_str::<Value>(text.trim()) {
+ Ok(value) => value,
+ Err(_) => return false,
+ };
+ &parsed
+ } else {
+ value
+ };
+ let Some(object) = value.as_object() else {
+ return false;
+ };
+ if object.get("success").and_then(Value::as_bool) == Some(true) {
+ return false;
+ }
+ match object.get("error") {
+ Some(Value::Null | Value::Bool(false)) | None => false,
+ Some(Value::String(message)) => !message.trim().is_empty(),
+ Some(_) => true,
+ }
+ }
+
+ update.get("rawOutput").is_some_and(explicit_error)
+ || update
+ .get("content")
+ .and_then(|content| content_text(Some(content)))
+ .is_some_and(|text| explicit_error(&Value::String(text)))
+}
+
+fn finish_tool(output: &ParentOutput, agent_id: &str, id: &str, update: &Value) {
+ let is_error = tool_result_is_error(update);
let mut result = tool_result_text(update);
if result.trim().is_empty() {
// Some adapters report completion with neither content nor rawOutput; a
@@ -2429,7 +2481,10 @@ fn finish_tool(output: &ParentOutput, id: &str, update: &Value) {
output.send(json!({
"type": "tool_execution_end",
"toolCallId": id,
+ "agentId": agent_id,
"toolName": tool_name(update),
+ "kind": tool_kind(update),
+ "args": tool_args(update),
// The desktop event router reads the raw-Pi result shape
// ({content: [{text}]}), not a bare string.
"result": { "content": [{ "type": "text", "text": result }] },
@@ -5533,6 +5588,17 @@ mod tests {
.expect("output progress");
assert_eq!(progress["outputDelta"], json!("compiling...\n"));
+ // Late ACP metadata must be visible to the desktop before completion.
+ let progress = tool_progress(&json!({
+ "title": "mcp__screenpipe__search-content",
+ "kind": "search",
+ "rawInput": { "query": "late metadata" }
+ }))
+ .expect("metadata progress");
+ assert_eq!(progress["title"], json!("mcp__screenpipe__search-content"));
+ assert_eq!(progress["kind"], json!("search"));
+ assert_eq!(progress["args"], json!({ "query": "late metadata" }));
+
// A bare status merge carries nothing renderable.
assert_eq!(tool_progress(&json!({ "status": "in_progress" })), None);
}
@@ -5682,6 +5748,50 @@ mod tests {
assert_eq!(events_of_type(&events, "tool_execution_end").len(), 1);
}
+ #[test]
+ fn terminal_tool_update_keeps_metadata_and_error_payload() {
+ let output = ParentOutput::buffer();
+ let state = test_state(&output);
+ state.handle_update(json!({
+ "sessionUpdate": "tool_call",
+ "toolCallId": "late-tool",
+ "title": "MCP: tool",
+ "status": "in_progress",
+ "rawInput": {}
+ }));
+ output.drain();
+
+ state.handle_update(json!({
+ "sessionUpdate": "tool_call_update",
+ "toolCallId": "late-tool",
+ "title": "mcp__screenpipe__search-content",
+ "kind": "search",
+ "status": "completed",
+ "rawInput": { "query": "late ACP metadata" },
+ "rawOutput": { "error": "Tool execution error" }
+ }));
+ let events = output.drain();
+ let ends = events_of_type(&events, "tool_execution_end");
+ assert_eq!(ends.len(), 1);
+ assert_eq!(ends[0]["agentId"], json!("test-agent"));
+ assert_eq!(
+ ends[0]["toolName"],
+ json!("mcp__screenpipe__search-content")
+ );
+ assert_eq!(ends[0]["kind"], json!("search"));
+ assert_eq!(ends[0]["args"], json!({ "query": "late ACP metadata" }));
+ assert_eq!(ends[0]["isError"], json!(true));
+
+ assert!(!tool_result_is_error(&json!({
+ "status": "completed",
+ "rawOutput": { "success": true, "error": "diagnostic field" }
+ })));
+ assert!(!tool_result_is_error(&json!({
+ "status": "completed",
+ "rawOutput": { "error": null }
+ })));
+ }
+
#[test]
fn advertises_and_detects_the_subagent_transcript_capability() {
let meta = subagent_transcript_capability();
diff --git a/docs/pr-assets/acp-tool-metadata/after.png b/docs/pr-assets/acp-tool-metadata/after.png
new file mode 100644
index 000000000..74ba48d9b
Binary files /dev/null and b/docs/pr-assets/acp-tool-metadata/after.png differ
diff --git a/docs/pr-assets/acp-tool-metadata/before.png b/docs/pr-assets/acp-tool-metadata/before.png
new file mode 100644
index 000000000..ed43ef496
Binary files /dev/null and b/docs/pr-assets/acp-tool-metadata/before.png differ