Thanks for opening this — it has been seen, and it is queued.

This note is automated, but it is not a brush-off: it exists so you know where your PR stands instead of having to guess from silence.

Current review status: working through a backlog. 0.9.1-rc.1 is out, so the release freeze that held reviews is over — but it left a large queue of open pull requests behind it, and we are reading through them oldest-first. The background is in discussion #1144.

What that means for this PR, concretely:

• It will not be closed for inactivity. No stale bot touches pull requests here.
• It may still sit a while before a human reads it. That is on us, not on you.
• Older PRs are read first, so a recent one is not being skipped — it is behind a queue.

Things that will genuinely speed it up whenever review does happen:

• Keep it rebased on main — the tree is moving quickly right now, and a conflicting branch cannot be reviewed as the diff you intended.
• Get CI green, or say which failures you believe are pre-existing.
• Keep the change to one claim. Bundled features and refactors get split before they get merged, which costs you a round trip.
• Every commit needs a sign-off (git commit -s) — CI enforces DCO.

If this fixes a bug, a reproduction we can run is worth more than a description of the symptom.

Thanks for contributing, and sorry in advance for the wait.

Could a maintainer please rerun the failed test-windows-guards job?

The failure occurred during setup in test_daemon_stability.py:

SETUP FAIL: permanent daemon did not start for the crash check

The preceding daemon lifecycle and stability sections passed. Several following guards then failed or skipped on daemon/index/UI startup preconditions. This PR only changes the Windows search_code PowerShell producer cap, so the daemon startup failure appears unrelated and potentially transient.

Maintainer decision on this one, and it goes a different way — but the problem you identified is real and is going to get solved properly.

We will not take a producer-side cap. The reason is result completeness. Our C-side cap counts accepted results: at src/mcp/mcp.c:9245 a line that fails the path_filter does continue without spending quota, so today PowerShell keeps streaming until we have 500 matching results. Select-Object -First 500 stops it at 500 unfiltered lines, and we then filter those down. On an unscoped search with a path_filter on Windows that silently returns somewhere between 0 and 500 where it previously returned 500 — and total_grep_matches reports the truncated count as though it were the whole answer.

Two further consequences: the same query would return different results on Windows and POSIX, and it re-establishes the producer as the truncation authority, which the comment at mcp.c:9770 records as deliberately removed (the -m flag let large vendored files exhaust the quota before reaching project source).

In this project a search that quietly returns fewer results than it found is a correctness bug, not a performance tradeoff. So a bound is right — a silent one is not.

The direction instead: generic pagination for search_code, on every platform. The response carries a page indicator and a signal for whether more results exist, and a flag lets the caller page through. That solves what you are actually solving — bounding a pathological scan — without any platform behaving differently from another and without ever discarding a result the caller cannot know about. There is precedent in the codebase: tools/list is already cursor-paginated (MCP_TOOLS_PAGE_SIZE).

So this PR will be closed in favour of that work, not because the diagnosis was wrong but because the bound belongs at a different layer.

What I would genuinely like, if you are up for it: #1604 is being taken — it is results-preserving by construction and brings Windows to the parity POSIX already has via grep --include. And your #1606 is already merged; the static char buf in cbm_mkdtemp was a real concurrency defect where two daemon sessions could be handed the same temp directory, and it was the cleanest PR of the five.

One request for the cluster generally: several of these PRs re-wrap comments in code they do not otherwise change, including in unrelated tests. It inflates review and creates conflicts between your own sibling PRs — #1604, #1607 and #1608 all rewrite the same snprintf block and all rename build_grep_cmd, so whichever lands second collides. Worth a formatter setting matched to the repo's column width.

Thank you for #1565 and for putting five concrete PRs behind it. The pathological-scan problem is real and it is now on the roadmap with a shape we can support long-term.

Closing in favor of the maintainer's planned cross-platform search_code pagination. Agreed that a producer-side row cap can silently underfill path-filtered results and report an incomplete count as complete.