Skip to content

fix(harness): preserve scoped history and Terminal focus [SAP-3292] #1040

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
ynadge wants to merge 1 commit into from
Closed

fix(harness): preserve scoped history and Terminal focus [SAP-3292] #1040

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions .changeset/assistant-history-scope-and-focus.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
---
"@sapiom/harness": patch
---

Preserve scoped mixed history, per-workspace refresh ordering, and explicit Terminal focus.
6 changes: 6 additions & 0 deletions packages/harness/docs/assistant-history.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,3 +207,9 @@ Visibility changes publish on the existing Assistant state revision clock, and
an older read cannot overwrite newer completed evidence. An unreadable index
blocks discovery and execution while exact authorized End remains available;
previously verified pending children remain private during that failure.

History groups Terminal and Assistant records only for the same Studio and cwd.
Transcript-only Terminal records remain reachable, and explicit Terminal focus
outranks an earlier Assistant reveal. Assistant history tracks failures and newer
results per cwd; another directory's refresh cannot erase or roll them back.
A dormant Terminal does not label the Studio ended before lifecycle state arrives.
313 changes: 312 additions & 1 deletion packages/harness/web/e2e/assistant-history.spec.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -658,7 +658,7 @@ test("resumed Assistant yields to newer foreground Terminal work", async ({
for (const revisit of [false, true]) {
test(`Resume keeps the original revision after its bus commit and lost ACK${revisit ? " across Close and Back" : ""}`, async ({
page,
}) => {
}, info) => {
const probe = await setup(page);
probe.holdResume = true;
await review(page);
Expand All @@ -681,6 +681,10 @@ for (const revisit of [false, true]) {
await expect(page.getByRole("alert")).toContainText(
"Retry to check the same operation",
);
if (!revisit)
await page.screenshot({
path: info.outputPath("retry-resume-original-request.png"),
});
if (revisit) {
await page
.getByTestId("assistant-history-pane")
Expand Down Expand Up @@ -721,3 +725,310 @@ test("a definite Resume conflict permits a fresh availability check without auto
"Saved answer assistant-only",
);
});

const historyUnavailable = (page: Page) =>
page.getByText("Assistant history is unavailable. Reopen history to retry.", {
exact: true,
});
const refreshDirectory = (page: Page, directory: string) =>
publish(page, {
type: "session.status",
session: {
id: "sess-boot",
cwd: directory,
title: "Directory history check",
harness: "claude-code",
agentSessionId: "unlisted-native",
boundWorkflowPath: null,
status: "exited",
ready: false,
createdAt: "2026-01-01T00:00:00.000Z",
lastActiveAt: "2026-01-01T00:00:00.000Z",
},
});
const otherCwd = "/Users/demo/rfq-agent";

test("a successful directory refresh preserves another directory's history failure until that directory succeeds", async ({
page,
}, info) => {
await setup(page);
let failed = true;
const requests: string[] = [];
await page.route("**/api/sessions/assistant-history?**", (route) => {
const directory = new URL(route.request().url()).searchParams.get("cwd")!;
requests.push(directory);
return route.fulfill(
directory === cwd && failed
? { status: 503, json: {} }
: { json: { entries: directory === cwd ? [entry("recovered")] : [] } },
);
});
await rows(page);
await expect(historyUnavailable(page)).toBeVisible();
const before = requests.filter((dir) => dir === otherCwd).length;
await refreshDirectory(page, otherCwd);
await expect
.poll(() => requests.filter((dir) => dir === otherCwd).length)
.toBeGreaterThan(before);
await expect(page.getByText("Loading…", { exact: true })).toHaveCount(0);
await expect(historyUnavailable(page)).toBeVisible();
await historyUnavailable(page).scrollIntoViewIfNeeded();
await page.screenshot({
path: info.outputPath("history-directory-unavailable.png"),
});
failed = false;
await refreshDirectory(page, cwd);
await expect(page.getByTestId("assistant-history-recovered")).toBeVisible();
await expect(historyUnavailable(page)).toHaveCount(0);
await page
.getByTestId("assistant-history-recovered")
.scrollIntoViewIfNeeded();
await page.screenshot({
path: info.outputPath("history-directory-recovered.png"),
});
});

for (const oldResult of ["failure", "old metadata"] as const) {
test(`a slow multi-directory batch cannot restore ${oldResult} over a newer directory result`, async ({
page,
}) => {
await setup(page);
// Make Terminal discovery immediate; this test holds only the Assistant
// directory fan-out and must allow its first cwd request to settle.
await page.evaluate(async () => {
const url = performance
.getEntriesByType("resource")
.find(
(entry) => new URL(entry.name).pathname === "/src/lib/api.ts",
)!.name;
const { MockApi } = await import(url);
MockApi.prototype.sessionHistory = async () => [];
});
const firstResponse = page.waitForResponse(
(response) =>
new URL(response.url()).pathname ===
"/api/sessions/assistant-history" &&
new URL(response.url()).searchParams.get("cwd") === cwd,
);
let acmeRequests = 0;
let held: Route | null = null;
await page.route("**/api/sessions/assistant-history?**", (route) => {
const directory = new URL(route.request().url()).searchParams.get("cwd");
if (directory === otherCwd && held === null) {
held = route;
return;
}
if (directory !== cwd) return route.fulfill({ json: { entries: [] } });
acmeRequests++;
return route.fulfill(
acmeRequests === 1 && oldResult === "failure"
? { status: 503, json: {} }
: {
json: {
entries: [
{
...entry("ordered"),
title:
acmeRequests === 1
? "Old saved history"
: "Fresh saved history",
},
],
},
},
);
});
await rows(page);
await expect.poll(() => held !== null && acmeRequests === 1).toBe(true);
await (await firstResponse).finished();
await page.evaluate(
() =>
new Promise<void>((resolve) =>
requestAnimationFrame(() => requestAnimationFrame(() => resolve())),
),
);
await refreshDirectory(page, cwd);
await expect(page.getByTestId("assistant-history-ordered")).toContainText(
"Fresh saved history",
);
await held!.fulfill({ json: { entries: [] } });
await expect(page.getByText("Loading…", { exact: true })).toHaveCount(0);
await expect(page.getByTestId("assistant-history-ordered")).toContainText(
"Fresh saved history",
);
await expect(historyUnavailable(page)).toHaveCount(0);
});
}

test("a mixed Assistant row retains its transcript-only Terminal record without starting either engine", async ({
page,
}, info) => {
const probe = await setup(page);
await page.evaluate(async () => {
const url = performance
.getEntriesByType("resource")
.find(
(entry) => new URL(entry.name).pathname === "/src/lib/api.ts",
)!.name;
const { MockApi } = await import(url);
const history = MockApi.prototype.sessionHistory,
record = MockApi.prototype.sessionRecord;
MockApi.prototype.sessionHistory = async function (cwd: string) {
const rows = await history.call(this, cwd);
return cwd === "/Users/demo/acme-app"
? [
...rows,
{
harnessSessionId: "assistant-only",
agentSessionId: "retained-terminal-native",
harness: "claude-code",
cwd,
title: "Retained Terminal record",
lastActiveAt: "2026-09-14T00:00:00.000Z",
source: "transcript",
resumeMode: "agent-resume",
},
]
: rows;
};
MockApi.prototype.sessionRecord = async function (id: string) {
if (id !== "assistant-only") return record.call(this, id);
return {
...(await record.call(this, "2b6d9e10-7711-4c2a-8b0a-9e4f2d1c5a33")),
harnessSessionId: id,
mergedSessionIds: [id],
agentSessionId: "retained-terminal-native",
};
};
});
await rows(page);
await expect(
page.getByTestId("assistant-history-assistant-only"),
).toContainText("Terminal + Assistant");
await expect(
page.getByTestId("history-retained-terminal-native"),
).toHaveCount(0);
await page.getByTestId("assistant-history-assistant-only").click();
await expect(page.getByTestId("assistant-transcript")).toContainText(
"Saved answer assistant-only",
);
await page.screenshot({
path: info.outputPath("mixed-assistant-record.png"),
});
await page
.getByRole("button", { name: "View Terminal history", exact: true })
.click();
await expect(page.getByTestId("past-session-pane")).toContainText(
"Retained Terminal record",
);
await expect(page.getByTestId("session-transcript")).toContainText(
"Wire the screening webhook",
);
expect(probe.calls).toEqual([]);
await page.screenshot({ path: info.outputPath("mixed-terminal-record.png") });
});

test("Open Terminal from history outranks an earlier Assistant Resume focus", async ({
page,
}) => {
const probe = await setup(page);
await review(page);
await checkResume(page).click();
await resumeButton(page).click();
await expect(page.getByText("Paused", { exact: true })).toBeVisible();
await review(page);
await page
.getByRole("button", { name: "View Terminal history", exact: true })
.click();
await page.getByRole("button", { name: "Terminal", exact: true }).waitFor();
await page.evaluate(
() =>
new Promise<void>((resolve) =>
requestAnimationFrame(() => requestAnimationFrame(() => resolve())),
),
);
await expect(
page.getByRole("button", { name: "Terminal", exact: true }),
).toHaveAttribute("aria-pressed", "true");
await expect(page.getByTestId("dead-session-pane")).toBeVisible();
expect(probe.resumes).toHaveLength(1);
});

test("a resumed dormant Terminal is not labelled as an ended Studio before its lifecycle bus arrives", async ({
page,
}) => {
const probe = await setup(page);
probe.holdResume = true;
await review(page);
await checkResume(page).click();
await resumeButton(page).click();
await expect.poll(() => probe.resumes.length).toBe(1);
const value = probe.resumed("assistant-only");
await probe.resumes[0]!.fulfill({
json: {
...value,
session: { ...value.session, terminalState: "not-started" },
},
});
await expect(page.getByText("Paused", { exact: true })).toBeVisible();
await expect(page.getByText("Session ended", { exact: true })).toHaveCount(0);
await expect(
page.getByText("Nothing generated yet", { exact: true }),
).toBeVisible();
await publish(page, {
type: "assistant.state",
snapshot: {
hostInstanceId: "history-host",
authorityRevision: "account-a",
revision: 2,
enabled: true,
sessions: [],
lifecycles: [{ ...entry("assistant-only").lifecycle, revision: 4 }],
},
});
await expect(page.getByText("Session ended", { exact: true })).toBeVisible();
});

test("a Terminal summary in another directory remains separate from an Assistant with the same Studio ID", async ({
page,
}) => {
await setup(page);
await page.evaluate(async () => {
const url = performance
.getEntriesByType("resource")
.find(
(entry) => new URL(entry.name).pathname === "/src/lib/api.ts",
)!.name;
const { MockApi } = await import(url);
const history = MockApi.prototype.sessionHistory;
MockApi.prototype.sessionHistory = async function (cwd: string) {
const rows = await history.call(this, cwd);
return cwd === "/Users/demo/rfq-agent"
? [
...rows,
{
harnessSessionId: "assistant-only",
agentSessionId: "different-directory-native",
harness: "codex",
cwd,
title: "Other directory Terminal",
lastActiveAt: "2026-09-14T00:00:00.000Z",
source: "transcript",
resumeMode: "agent-resume",
},
]
: rows;
};
});
await rows(page);
await expect(
page.getByTestId("assistant-history-assistant-only"),
).not.toContainText("Terminal + Assistant");
await expect(
page.getByTestId("history-different-directory-native"),
).toBeVisible();
await page.getByTestId("assistant-history-assistant-only").click();
await expect(
page.getByRole("button", { name: "View Terminal history", exact: true }),
).toHaveCount(0);
});
17 changes: 11 additions & 6 deletions packages/harness/web/src/App.tsx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -749,6 +749,10 @@ export const App = (): JSX.Element => {
}, []);
const activeAssistantReview = assistantReview?.authority === harness.assistantHistoryAuthority &&
assistantReview.navigation === studioRestoreGenerationRef.current ? assistantReview : null;
const reviewedAssistantTerminal = activeAssistantReview ? harness.state?.sessions.find((session) =>
session.id === activeAssistantReview.entry.harnessSessionId && session.cwd === activeAssistantReview.entry.cwd) : undefined;
const reviewedTerminalSummary = activeAssistantReview ? harness.history.find((summary) =>
summary.harnessSessionId === activeAssistantReview.entry.harnessSessionId && summary.cwd === activeAssistantReview.entry.cwd) : undefined;
Comment on lines +752 to +755

@devin-ai-integration devin-ai-integration Bot Sep 14, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Aliased workspaces split mixed history

For symlinked workspaces, reviewedAssistantTerminal rejects the Terminal belonging to the reviewed Assistant. AssistantHistory.list returns realpath paths, while sessions retain their launch spelling. The mixed row loses its Terminal action and appears as separate history.

Learn more

Assistant history canonicalizes every requested and returned cwd with realpath. Terminal sessions retain the path normalized by path.resolve, which does not resolve symlinks. The new exact string comparisons therefore treat two spellings of the same directory as different scopes across both the rail and review pane.

Example: A session starts in /work/current, where current is a symlink to /repos/acme. Its Terminal session keeps /work/current, but its Assistant entry uses /repos/acme. Opening that Assistant shows no Terminal action, and history displays two rows instead of one mixed row.

Recommended fix: Give both history sources the same server-issued canonical cwd before comparing them. Apply the same identity rule in reviewedAssistantTerminal, reviewedTerminalSummary, assistantByStudio, and the row-level Terminal lookup; browser-only lexical normalization cannot resolve symlinks.

Devin Review

Was this helpful? React with 👍 or 👎 to provide feedback.

// Template gallery opened from the command palette (browse is reachable
// from anywhere, not only the add dialog / welcome panel entries).
const [templatesOpen, setTemplatesOpen] = useState(false);
Expand Down Expand Up @@ -3430,11 +3434,11 @@ export const App = (): JSX.Element => {
return true;
}}
/></>}
terminalNotStarted={state.sessions.find((session) => session.id === activeAssistantReview.entry.harnessSessionId)?.terminalState === "not-started"}
terminalLabel={state.sessions.find((session) => session.id === activeAssistantReview.entry.harnessSessionId)?.terminalState === "not-started" || state.sessions.find((session) => session.id === activeAssistantReview.entry.harnessSessionId)?.status !== "exited" ? "Open Terminal" : "View Terminal history"}
onOpenTerminal={state.sessions.some((session) => session.id === activeAssistantReview.entry.harnessSessionId)
? () => openSession(activeAssistantReview.entry.harnessSessionId)
: undefined}
terminalNotStarted={reviewedAssistantTerminal?.terminalState === "not-started"}
terminalLabel={reviewedAssistantTerminal && (reviewedAssistantTerminal.terminalState === "not-started" || reviewedAssistantTerminal.status !== "exited") ? "Open Terminal" : "View Terminal history"}
onOpenTerminal={reviewedAssistantTerminal
? () => { harness.revealTerminal(reviewedAssistantTerminal.id); openSession(reviewedAssistantTerminal.id); }
: reviewedTerminalSummary ? () => reviewPastSession(reviewedTerminalSummary) : undefined}
/>
) : showReview && reviewSummary ? (
<PastSessionPane
Expand Down Expand Up @@ -3930,7 +3934,8 @@ export const App = (): JSX.Element => {
source={canvasSource}
loadWorkflowGraph={shellApi.getWorkflowGraph.bind(shellApi)}
overviewActive={showComposer}
sessionExited={showDead && !!conversationSession && !assistantNeedsEnd(conversationSession.id)}
sessionExited={showDead && !!conversationSession && !assistantNeedsEnd(conversationSession.id) &&
(conversationSession.terminalState !== "not-started" || assistantLifecycles.some((row) => row.harnessSessionId === conversationSession.id && row.lifecycle === "ended"))}
onCanvasState={(hasContent) => {
// The board keeps its mount behind the map; its probe must
// not reveal or collapse the pane while the PROJECT is what
Expand Down
Loading
Loading