fix(server): skip live git in fetch_agent_history placement

[yinaoxiong]

Linked issue

Closes #1382

Type of change

• Bug fix

What does this PR do

The Sessions (history) list resolves every agent's placement through getCheckoutStatus(cwd), which spawns git subprocesses per cwd. On slow / network filesystems each git status takes seconds, and the sum across all distinct cwds blows past the client's 10s fetch_agent_history RPC timeout — so the list hangs and renders empty.

The history path doesn't use that git data at all: the client requests it with showCheckoutInfo={false}, never reads the project/checkout payload, and applies no project filter here. So the git work is pure overhead.

This PR makes fetch_agent_history resolve placement with refreshGit: false (cached snapshot + workspace registry lookup) instead of live git. The fetch_agents (active list) path is unchanged, the wire protocol is unchanged, and no client/CLI changes are needed.

How did you verify it

Tested on a real machine with ~180 sessions across 15 distinct cwds, several on network mounts.

Before (current code), from daemon.log:

• fetch_agent_history_request latency: 10.7s – 11.9s — over the 10s client timeout, list came back empty.

Micro-benchmark of the placement step (real getCheckoutStatus against the real cwds, vs the registry lookup the fix uses):

• Before (per-cwd git): ~21,000 ms total for 15 cwds (cold cache — the worst case the slow-FS user hits). Top offenders: one repo 4.6s, four network-mount repos ~3s each.
• After (registry lookup): ~0.03 ms.

End-to-end after the fix: built the server, swapped it into my daemon, restarted, and opened the Sessions list from the app — it now loads sub-second instead of timing out.

Automated test: added a test that mocks getCheckoutStatus to throw and asserts the history path still returns entries without ever calling it (proves no git is spawned). Full session.workspaces.test.ts passes (53 passed, 4 pre-existing skips).

Checklist

• One focused change. Unrelated cleanups split out.
• npm run typecheck passes (0 errors, all workspaces)
• npm run lint passes (oxlint: 0 warnings, 0 errors)
• npm run format ran
• UI changes include screenshots — N/A, server-only change
• Tests added or updated where it made sense

[greptile-apps]

Greptile Summary

This PR fixes a timeout in fetch_agent_history_request by passing { refreshGit: false } when resolving each session's project placement, replacing per-cwd live git status calls with a cached snapshot + registry lookup on the history path.

• session.ts: Introduces skipGitPlacement (true only for fetch_agent_history_request) and threads { refreshGit: false } into buildProjectPlacementForCwd on that path, leaving the active-list path (fetch_agents_request) unchanged.
• session.workspaces.test.ts: Adds a regression test that wires getCheckout to throw and asserts the history response still returns populated entries, proving no git subprocess is spawned.

Confidence Score: 5/5

Safe to merge — the change is two production lines scoped precisely to the history path, backed by a test that would fail if git is ever invoked on that path.

The fix is surgical: one boolean variable and one conditional argument. The active-list path is entirely unaffected. The { refreshGit: false } option already existed and was used in forwardAgentUpdate, so no new behavior surface is introduced. The fallback in resolveWorkspaceDirectory preserves existing null-return behavior.

No files require special attention.

Important Files Changed

Filename	Overview
packages/server/src/server/session.ts	Two-line targeted fix: adds skipGitPlacement scoped strictly to the fetch_agent_history_request branch and passes { refreshGit: false } to buildProjectPlacementForCwd. Active-list path is unaffected.
packages/server/src/server/session.workspaces.test.ts	New regression test wires getCheckout to throw, runs fetch_agent_history_request, and asserts the response contains the expected entry via ports-and-adapters injection.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[listFetchAgentsEntries] --> B{request.type?}
    B -->|fetch_agent_history_request| C[skipGitPlacement = true]
    B -->|fetch_agents_request| D[skipGitPlacement = false]
    C --> E[getPlacement with refreshGit: false]
    D --> F[getPlacement with undefined options]
    E --> G[resolveWorkspaceDirectory]
    F --> G
    G --> H{refreshGit === false?}
    H -->|Yes - history path| I[peekSnapshot - cached\nno git subprocess]
    H -->|No - active path| J[getCheckout - live git]
    I --> K[workspace registry lookup]
    J --> K
    K --> L[ProjectPlacementPayload]

Loading

_{Reviews (2): Last reviewed commit: "fix(server): skip live git in fetch_agen..." | Re-trigger Greptile}

[greptile-apps]

Mock call-count assertion on the boundary

expect(getCheckout).not.toHaveBeenCalled() asserts implementation internals rather than observable behavior. The test already covers the behavioral goal because getCheckout is wired to throw — if the history path ever calls it, the test fails with an unhandled rejection before reaching the expect(emitted) check. The explicit not-called assertion is therefore redundant and falls into the pattern the project rules flag (asserting spy/mock call counts instead of recording observable state). Consider relying solely on the throw-on-call guard and the emitted-entries assertion.

Rule Used: # Code Review Pattern Reference: Slop, Tests, Feat... (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!