fix(docs-hygiene): compress caveman/taxonomy audit remediations (0.15.0)

[kyle-sexton]

Closes #2745

Summary

Replacement for #2805 (auto-closed when stack base deleted). Compress caveman/taxonomy audit remediations on top of #2810.

Related

• Top of docs-hygiene stack after docs(docs-hygiene): process derivability route-to-sibling batch (0.14.7) #2810 (0.14.7).
• Supersedes closed fix(docs-hygiene): compress caveman/taxonomy audit remediations (0.15.0) #2805.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[claude]

Claude finished @kyle-sexton's task in 1m 4s —— View job

---

Security review (/review:security-review)

• Gather context (diff, changed files, CLAUDE.md)
• Invoke /review:security-review skill and follow its procedure
• Post findings / results

Scope reviewed: full diff origin/cursor/2735-route-sibling-8f91...HEAD (13 files, +313/-53) — plugins/docs-hygiene/ skill docs (SKILL.md, context/*.md), CHANGELOG.md, plugin.json version bump, evals/evals.json + new fixture, and two shell scripts (scripts/audit-scan.sh new, scripts/detect-caveman.sh modified) plus their test scripts.

No security issues found.

This PR is documentation/tooling for the docs-hygiene:compress skill — prose-compression methodology docs, eval fixtures, and local dev-tooling shell scripts. It has no security-relevant surface:

• No GitHub Actions workflow files are touched (no permissions:, triggers, or CI changes).
• No authentication, authorization, secrets, tokens, or network I/O are involved.
• The two shell scripts (audit-scan.sh, detect-caveman.sh) operate only on local file paths supplied by the invoking session/skill (the same trust boundary as the user's own repo) — no eval, no unsanitized interpolation into a shell context, filenames handled via quoted arrays/variables throughout. detect-caveman.sh's jq usage takes JSON via a shell variable (printf '%s' "$json" | jq ...), not string-interpolated into the jq filter itself, so it's not injectable.
• SKILL.md's revised Step B (temp-dir mechanical-compression backend) still confines writes to a mktemp -d tempdir and only copies back over the real target on explicit success — same safety property as before, just restructured for the "Bash state doesn't persist across tool calls" constraint.

Per the security-review skill's high-signal bar, no pre-existing/style/lint findings are reported here — see /review:code-review for that lane.

[github-actions]

Last security-reviewed head: afe41531fca7d8ab6cef6e20eb33853b9df4732d. On the next push, the relevance gate compares only the commits since this SHA; delete this comment to force a full re-review.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 49588fefcf

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[claude]

Claude finished @kyle-sexton's task in 1m 34s —— View job

---

Security review (/review:security-review)

• Gather context (diff, changed files, CLAUDE.md)
• Invoke /review:security-review skill and follow its procedure
• Post findings / results

Scope reviewed: full diff origin/cursor/2735-route-sibling-8f91...HEAD at head afe4153 (13 files, +324/-53) — plugins/docs-hygiene/ skill docs (SKILL.md, context/*.md), CHANGELOG.md, plugin.json version bump, evals/evals.json + new fixture, and the two shell scripts (scripts/audit-scan.sh, scripts/detect-caveman.sh) plus their test scripts.

No security issues found.

This is documentation/tooling for the docs-hygiene:compress skill with no security-relevant surface:

• No GitHub Actions workflow files are touched (no permissions:, triggers, or CI changes).
• No authentication, authorization, secrets, tokens, or network I/O are involved.
• scripts/audit-scan.sh (permalink) operates only on local .md paths passed as arguments — quoted throughout, no eval, no unsanitized interpolation into a shell/exec context. grep -E/-o patterns are static literals, not built from file content.
• scripts/detect-caveman.sh (permalink) pipes claude plugin list --json output into jq via a shell variable (printf '%s' "$json" | jq -r '...') with a fixed filter string, not string-interpolated JSON into the filter — not injectable. The added enabled == true / pinned-id-preference logic is a correctness change, not a trust-boundary one.
• SKILL.md's revised Step B (temp-dir mechanical-compression backend) still confines writes to a mktemp -d tempdir and only copies back over the real target on explicit success — same safety property as before, restructured only for the "Bash state doesn't persist across tool calls" constraint.
• Diffed the incremental commit since the prior review pass (01726f9 → afe4153, the "satisfy markdownlint/typos/shellcheck" fixup): only comment/whitespace additions and an unused-variable removal in audit-scan.sh — no new security-relevant surface.

Per the security-review skill's high-signal bar, no style/lint/correctness findings are reported here (two P2 detection-accuracy items were already raised by the automated Codex review on audit-scan.sh — those are heuristic-correctness, not security, so out of scope for this lane). See /review:code-review for that lane.

[claude]

Claude encountered an error —— View job

---

I'll analyze this and get back to you.

[github-actions]

Warning

Automated review did not complete — this is an infrastructure failure, not a review verdict.

Treat any Claude comment on this PR (including a placeholder like "I'll analyze this and get back to you") as incomplete, not "no findings."

• Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/31919113872
• Failure class: other
• Last SDK result: (no execution file was produced)

Re-run the job, or workflow_dispatch this workflow with the PR number, to retry the review. A new push re-triggers this lane only if the caller's pull_request triggers include synchronize (the canonical caller omits it).
An automatic retry may already have run — it is skipped when a partial review could duplicate comments, or when the failure class needs an operator (auth).

[claude]

Claude encountered an error —— View job

---

I'll analyze this and get back to you.