Repository navigation
Description
Problem
Every informational comment disposition dev-lead posts in a fleet repo is rejected by the harness, so the original comment is never minimized RESOLVED and pr-review's undispositioned-pr-comment gate holds the PR indefinitely.
cdv_body_has_findings (scripts/lib/comment-disposition-verify.sh:505) resolves its own directory lazily, at call time:
lib_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"dev-lead-fix-reviews.sh is run as bash .dev-lead/scripts/dev-lead-fix-reviews.sh, so BASH_SOURCE[0] is the relative path .dev-lead/scripts/lib/comment-disposition-verify.sh. By the time the function is called (dev-lead-fix-reviews.sh:1734) the script has already cd'd into the PR worktree (checkout_pr_in_worktree, line 118). The cd fails, lib_dir is empty, the registry cannot be sourced, and the function takes its fail-closed path and returns 0 ("has findings"). The disposition is then marked verified=false.
The script already knows about this hazard for its own paths (AGENT_SCRIPT_DIR is "captured before any cd into a PR worktree"); the library function was missed.
Evidence
Run 37983598107 (google-app-scripts#594, dev-lead v139-stable, 2026-10-09 19:56Z):
.dev-lead/scripts/lib/comment-disposition-verify.sh: line 505: cd: .dev-lead/scripts/lib: No such file or directory
##[notice]comment IC_kwDORFz9wc8AAAABVtpOwA carries a finding-bearing section — an `informational` disposition cannot cover it; each finding needs a real disposition (#2008)
##[notice]skipping comment IC_kwDORFz9wc8AAAABVtpOwA — disposition 'informational' not authorized to resolve (is_human=false verified=false); leaving open (#1813)
IC_kwDORFz9wc8AAAABVtpOwA is gemini-code-assist's "Gemini is unable to generate a review for this pull request due to the file types involved not being currently supported." It has no finding-bearing section: pr-review's own gate classifier reports findings=false for it.
Local reproduction against main (0ae6227d), sourcing the library by a relative path:
cwd=root rc=1 # no findings → informational verifies
cwd=elsewhere rc=0 # "has findings" → informational rejected
comment-disposition-verify.sh: line 505: cd: .dev-lead/scripts/lib: No such file or directory
Impact
- The retry sweep reports these comments as
dispositionedand stops retrying, while pr-review's gate reports them as blocking. Nothing re-fires. - Open PRs currently held by at least one such comment: TalkTerm#509 (its only blocking comment), TalkTerm#521, markets#502, google-app-scripts#594. Any PR that receives a Gemini "unsupported file types" notice or a no-finding CodeRabbit summary is affected.
- Each retry spends a model run and ends
no-changes. - Present since CodeRabbit's Security Architecture Review is throttled separately from its code review — findings in both sections must be addressed, and an in-place edit after disposition must re-open the comment #2008 landed (2026-10-02).
Acceptance criteria
cdv_body_has_findingsreturns the same result regardless of the caller's working directory when the library was sourced by a relative path.- A regression test sources the library by a relative path, changes directory, and asserts a no-finding notice body returns 1.
- The fail-closed behaviour for a genuinely unreadable registry is unchanged.
Notes
scripts/lib/maintainer-comment-gate.sh lines 92 and 127 use the same lazy BASH_SOURCE pattern inside functions. No failure has been observed there (pr-review invokes it differently), so it is not in scope here.
Activity
github-actions commentedon Oct 9, 2026
Thanks for the detailed report and the clear reproduction — the evidence from run 37983598107 and the local cwd comparison makes the root cause easy to follow. One question to help us scope the fix: besides
dev-lead-fix-reviews.sh, are there any other callers that sourcecomment-disposition-verify.shby a relative path and thencdbefore callingcdv_body_has_findings, or is that script the only one affected?don-petry commentedon Oct 10, 2026
Second occurrence, on a different repo and ring:
petry-projects/.githubPR #1269 (feat: implement issue #1267), dev-leadv139-ring0atf6f2e384, run 38012232150, 2026-10-10 01:12Z.The comment in question is CodeRabbit's "Action not completed — Review rate limited" reply, which has no finding section at all. That PR is green with no open threads and is held only by
undispositioned-pr-commenton four bot comments, so it is waiting on this fix reachingdev-lead/v139-ring0.fix(dev-lead): resolve the disposition verifier's lib dir at source t…
feat: implement issue #2208 — dev-lead unminimizes every informationa…