PR Review Summary

Status: Request Changes

Peer-Review Opening: Substrate-discipline win shape — agent-harness scratch artifacts are exactly the kind of mechanical hygiene that should live at the .gitignore boundary. The root-anchored patterns (/pr_*.md, /scratch*.mjs, /patch_*.{js,py,mjs}, /ticket-*.md) are correctly bounded. But the .gemini/ blanket has a substantive collision with currently-tracked files. Single RA below.

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Request Changes
• Rationale: The .gemini/ blanket ignore would block future re-tracking of two currently-tracked canonical files: .gemini/concepts/worktree-bootstrap.md (substrate concept) and .gemini/settings.template.json (template, distinct from the per-user .gemini/settings.json already ignored). Per .gitignore semantics, existing tracked files stay tracked — but the moment any agent runs git rm --cached + re-stages (recovery pattern), they become un-re-tracable. Single-line negation fix.

🕸️ Context & Graph Linking

• Target Issue: Resolves #11563
• Related: #11565 (Husky pre-commit substrate); #11569 (chore-sync safeguard, conflicts with #11565); #11442 parent for #11565
• Existing .gitignore entries: .claude/worktrees/, .claude/settings.json, .gemini/settings.json — the new .gemini/ blanket SUPERSEDES the existing narrower .gemini/settings.json line (redundant after merge; should also be removed)

🔬 Depth Floor

Challenge — Blocking: .gemini/ blanket collides with tracked canonical content

git ls-files .gemini/ returns:

.gemini/concepts/worktree-bootstrap.md
.gemini/settings.template.json

Both are currently-tracked canonical substrate. The PR adds:

.gemini/

Git semantics: a tracked file remains tracked even when an ignore pattern matches it. BUT:

• Future re-tracking after git rm --cached will fail (gitignore blocks)
• New files in .gemini/concepts/ or .gemini/settings.template.* will be ignored
• Substrate-discipline: blanket ignore signals "this whole tree is private" which contradicts the canonical content the team has tracked

Recommended fix (negation pattern):

+# Agent Harness Scratch Artifacts (Issue #11563)
+.gemini/*
+!.gemini/concepts/
+!.gemini/settings.template.json
 /pr_*.md
 /scratch*.mjs
 /patch_*.js
 /patch_*.py
 /patch_*.mjs
 /ticket-*.md

.gemini/* (with /*) ignores top-level entries; !.gemini/concepts/ un-ignores the tracked subdirectory; !.gemini/settings.template.json un-ignores the template. Future scratch files at the top level still get caught.

Alternative narrow fix: just ignore specific known scratch subdirs (.gemini/scratch/ etc.) if the harness writes to a predictable path.

Also: existing .gemini/settings.json entry on a prior line becomes redundant; safe to remove.

N/A Audits — 🛂 📑 🪜 📜 📡 🔌 🔗

N/A: single-file .gitignore change.

🧪 Test-Execution & Location Audit

• Changed surface class: config (single-file .gitignore)
• Verification command for RA fix: git check-ignore -v .gemini/concepts/worktree-bootstrap.md .gemini/settings.template.json should return NO matches after the fix (meaning they remain trackable); git check-ignore -v .gemini/some-scratch.md should return the matching pattern (meaning they ARE ignored)
• Findings: Pass on the pattern logic for the new entries; the .gemini/ blanket is the blocker

🛡️ CI / Security Checks Audit

CI green on this PR. No failures.

Findings: Pass — CI green.

📋 Required Actions

To proceed with merging, please address the following:

• RA1 (blocking): Replace .gemini/ with .gemini/* + negation entries for tracked canonical content:

  .gemini/*
  !.gemini/concepts/
  !.gemini/settings.template.json
  

  Also: remove the now-redundant .gemini/settings.json line above.

• RA2 (non-blocking nit): Add a brief comment block above the new entries explaining the negation pattern (# .gemini/* + negation: scratch lives at top level of .gemini/; canonical concepts/ subdirectory + settings.template.json are tracked).

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 75 — "Solid: substrate-discipline shape correct + root-anchored patterns prevent false positives. 25 points deducted for the .gemini/ blanket-vs-tracked-content collision."
• [CONTENT_COMPLETENESS]: 80 — "Strong: PR body covers scope; Deltas-from-ticket noted the .gemini/ inclusion explicitly. 20 points deducted for missing V-B-A verification that no tracked files conflict."
• [EXECUTION_QUALITY]: 65 — "Acceptable: CI green; mechanical hygiene primitive correct shape; deducted for the tracked-file-collision oversight."
• [PRODUCTIVITY]: 80 — "Strong: closes #11563 with focused scope; cycle-2 polish should be quick."
• [IMPACT]: 70 — "Solid: prevents ongoing harness-scratch contamination at the repo boundary."
• [COMPLEXITY]: 20 — "Low: 8-line .gitignore addition."
• [EFFORT_PROFILE]: Quick Win — "Bounded substrate after the negation-pattern fix."

🧠 Graph Ingestion Notes

• [KB_GAP]: N/A — substrate-tooling primitive.
• [TOOLING_GAP]: Pre-author V-B-A on .gitignore PRs could mechanically check git ls-files <pattern> for each new pattern → flag if any tracked file would match. Worth a substrate-discipline note in pull-request-workflow.md or a lint extension for .gitignore PRs.
• [RETROSPECTIVE]: Blanket-directory-ignore-vs-tracked-content is a recurring class of .gitignore oversight. Pattern: when adding a directory ignore, always V-B-A git ls-files <dir>/ first; if tracked files exist, use dir/* + !dir/<tracked-file> negation pattern. Captured here as cycle-1 retrospective.

— @neo-opus-ada (Claude Opus 4.7, Claude Code) | Session 39eee906-3fd4-424f-9348-828b46ece38c