fix(ai-slop): close leftover directory-expansion defects (0.3.9)

[kyle-sexton]

Closes #3332

Summary

Three defects in detect.sh directory expansion survived the 0.3.8 path-anchor fix: a tracked non-ASCII filename was dropped, a filesystem walk still ran silently when git could not answer, and the new tests asserted only a scan count so the original spelling-rebuild defect could return unnoticed.

Fix

• git -c core.quotePath=false ls-files so a filename holding an em dash is actually opened.
• Report on stderr when git is missing or rev-parse --is-inside-work-tree fails; a silent walk remains only when git reports the directory is genuinely outside a checkout.
• Dir-target tests now assert the emitted file= path, subtree restriction, a non-ASCII filename, and the git-absent report.
• normalize_dir_target still keeps a Windows drive-root slash (C:/).

ai-slop 0.3.9. Duplicate stale PR #3330 is still DIRTY and would regress drive-root slash preservation; ignore it.

Verification

• bash plugins/ai-slop/skills/audit/scripts/detect.test.sh: 149 cases passed
• scripts/check-changed-skills.sh origin/main: 0 failed
• scripts/check-changelog-parity.sh --check-bump origin/main: pass
• python3 scripts/sync-plugin-options-docs.py --check: up to date

Related

Follow-up to the merged directory-anchor work. Do not merge #3330.

  

[claude]

Claude finished @kyle-sexton's task in 1m 8s —— View job

---

Security Review

• Gather context (PR diff, changed files)
• Run /review:security-review per custom instructions
• Post findings

No security issues found.

This PR is a narrow, local-tooling bug fix in plugins/ai-slop/skills/audit/scripts/detect.sh. It adds -c core.quotePath=false to an existing git ls-files '*.md' invocation inside expand_dir_target so tracked filenames containing non-ASCII bytes (e.g. an em dash) are emitted raw instead of C-quoted, fixing a case where such files were silently dropped from the audit's markdown scan.

Reviewed for the categories in scope for this lane:

• Command/path injection: dir and the *.md glob are passed as quoted arguments to git -C/ls-files, same as the pre-existing code; -c core.quotePath=false is a formatting-only git config value, not a code-execution or trust-boundary setting, and is not influenced by untrusted input beyond what the function already accepted.
• Trust boundary / data flow: no new external input source is introduced; the only behavioral change is how already-tracked repo filenames are decoded from git's own output.
• GitHub Actions surface: no workflow files, permissions, or triggers are touched.
• Test file (detect.test.sh): only adds a local fixture (a tracked file with an em-dash filename) and assertions; no injection or credential handling involved.

Nothing here meets the skip-gate bar for further work — no security-relevant surface in this diff.

Branch: cursor/aislop-quotepath-cfcf

[github-actions]

Last security-reviewed head: 0e6b30c97d2101b221f89cd36cfd5d810141332f. On the next push, the relevance gate compares only the commits since this SHA; delete this comment to force a full re-review.

[claude]

Claude finished @kyle-sexton's task in 1m 54s —— View job

---

Security Review

• Gather context (PR diff, changed files)
• Run /review:security-review per custom instructions
• Post findings

No security issues found.

This PR is a local-tooling bug fix in plugins/ai-slop/skills/audit/scripts/detect.sh's expand_dir_target (this PR's diff builds on the merged directory-anchor fix, closing three leftover directory-expansion defects). Reviewed the full PR diff against the categories in scope for this lane:

• Non-ASCII path handling (detect.sh#L323): adds -c core.quotePath=false to git ls-files so tracked filenames with non-ASCII bytes (e.g. an em dash) are emitted raw instead of C-quoted, fixing files that were silently dropped from the scan. core.quotePath only affects quoting of bytes ≥ 0x80; git always escapes control characters (including embedded newlines) and backslash/double-quote regardless of this setting, so the newline-delimited while IFS= read -r rel; do ... done <<<"$listing" loop at detect.sh#L330 is not newly exposed to line-splitting/injection via crafted filenames — that guarantee predates and is unaffected by this change.
• New git-absent / no-worktree fallback paths (detect.sh#L305-L317): both branches only change when the pre-existing find "$dir" -name '*.md' -type f fallback runs and add an stderr note; no new command construction or input source.
• Command/path injection: dir and the *.md glob remain quoted arguments to git -C/ls-files/find, unchanged from the prior version; -c core.quotePath=false is a formatting-only git config value, not code-execution or trust-boundary relevant.
• Test file (detect.test.sh): adds fixtures (an em-dash-named tracked file, a PATH stub without git built from ln -s to existing local binaries) and assertions only — no untrusted input or credential handling.
• No GitHub Actions, permissions, or workflow files touched.

No security-relevant surface introduced by this diff.

Branch: cursor/aislop-quotepath-cfcf

[claude]

Claude finished @kyle-sexton's task in 1m 12s —— View job

---

Security Review

• Gather context (PR diff, changed files)
• Run /review:security-review per custom instructions
• Post findings

Skip gate: no new security-relevant surface since the last successful review.

Diffing the previously-reviewed head (4c4dfe4) against the current head (0e6b30c) shows zero changes to plugins/ai-slop/** — the only delta is unrelated plugins/computer-use/** content that entered via a rebase onto main, not new commits on this PR. The ai-slop changeset (plugin.json, CHANGELOG.md, detect.sh, detect.test.sh) is byte-identical to what was already reviewed and cleared in the prior two passes on this PR.

No new review performed; the prior finding stands: no security issues found.

Branch: cursor/aislop-quotepath-cfcf

[claude]

Claude finished @kyle-sexton's task in 19s —— View job

---

Code Review

• Gather context (PR diff, changed files)
• Check /review:code-review skip gate
• Post findings

Skipped: PR is merged (not open).

Per /review:code-review's skip gate, review stops early when "PR is closed or not open," and no further findings are posted in that case. This PR (#3331) is already MERGED, so there's no open PR to review or anchor inline comments to — the code-review lane doesn't apply retroactively to merged history.

If you'd like a review of this code as it stands on main now, or of a new/reopened PR, just ask and I'll run the review lane against that target.

[github-actions]

Claude has reviewed this PR 1 time. The lane skips further automatic reviews after 5; deleting this comment resets the count.