dev-lead unminimizes every informational-dispositioned bot comment on its next pass: maintainer-comment-gate.sh resolves its lib dir after the cd into the PR worktree (#2152 again, second file) #2208

Problem

dev-lead minimizes a bot comment RESOLVED on one pass and unminimizes it on the next, so pr-review's undispositioned-pr-comment gate can never settle on a PR that needs an informational disposition.

scripts/lib/maintainer-comment-gate.sh resolves its own directory lazily, at call time, in two functions:

# _maintainer_gate_info_patterns_json (line 92) and _maintainer_gate_registry_value (line 127)
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/maintainer-comment-gate.sh. By the time resolve_dispositioned_comments calls maintainer_gate_reopen_candidates, the script has already cd'd into the PR worktree (checkout_pr_in_worktree). The cd fails, the registry cannot be sourced, and every lookup takes its fail-closed default:

resolved_verdict then returns block for any RESOLVED bot comment whose covering disposition is informational, and the harness unminimizes it. The stderr is discarded (2>/dev/null on the call), so the run log shows only the unminimize notice.

This is the same defect as #2152 (fixed for comment-disposition-verify.sh by #2154, which captures _CDV_LIB_DIR at source time). This file was missed.

Evidence

petry-projects/.github PR 1276 (fix(canary): fail closed on a non-array run list), dev-lead v139-ring0 at 6fbedea, 2026-10-10. Comments IC_kwDORyesfc8AAAABayritw (Codex usage-limit notice) and IC_kwDORyesfc8AAAABayrkCg (Qodo trial-ended notice) were never edited (lastEditedAt is null) and each has one trusted informational disposition posted at 11:37Z.

• Run 38055941317, 14:05Z: minimized comment IC_kwDORyesfc8AAAABayritw RESOLVED (disposition=informational is_human=false) and the same for …ABayrkCg.
• Run 38067423978, 16:24Z, same tag: unminimized comment IC_kwDORyesfc8AAAABayritw — edited after its latest disposition (or its disposition cannot cover a finding-bearing body) and the same for …ABayrkCg. Nothing about either comment had changed in between.

Reproduction on main at aef61164:

bash -c 'source scripts/lib/maintainer-comment-gate.sh; cd /tmp
         _maintainer_gate_finding_re; echo; _maintainer_gate_registered_logins_json; echo; _maintainer_gate_info_patterns_json'
# cd: scripts/lib: No such file or directory   (three times)
# [\s\S]
# null
# {}

Without the cd /tmp the same three calls return the registry pattern, a 9-element array and a 4-key object.

Acceptance criteria

• scripts/lib/maintainer-comment-gate.sh captures its library directory once, at source time, in a readonly variable (as _CDV_LIB_DIR does), and both functions above use it. No function in this file calls dirname "${BASH_SOURCE[0]}".
• A test sources the gate by a relative path, changes to an unrelated directory, and asserts: _maintainer_gate_finding_re equals reviewer_sources_finding_section_pattern (not [\s\S]), _maintainer_gate_registered_logins_json is a non-empty array, and _maintainer_gate_info_patterns_json is a non-empty object.
• A test asserts that, after the same directory change, maintainer_gate_reopen_candidates returns [] for a never-edited bot comment that is minimized RESOLVED, has a clean body, and is covered by one trusted informational disposition. The test fails on main before the fix.
• The fail-closed defaults stay as they are for a registry that is really missing (existing tests still pass).

Out of scope

scripts/maintainer-resolve-comment.sh (lines 187 and 449) uses the same call-time pattern but is run standalone, not from a PR worktree. Leave it unless a test shows it failing.

Related: #2152, #2154, #2008, #2166, #1813.

Filed from a Claude Code session diagnosing why pr-review does not review petry-projects/.github PR 1276, at the maintainer's request. All times UTC.