[kyle-sexton]

Closes #8737

Summary

The seven lane-permissions behavior-probes cases had never run live, so nothing recorded whether the write lane's Read deny rules still keep Grep, Glob and Read out of /proc/*/environ and the checkout's .git on a current Claude Code. This PR records their first live run.

Fix

• records.md adds record 17 and its basis: all seven cases pass on Claude Code 2.1.312 under dontAsk with the lane's --allowedTools and --disallowedTools.
• A bump: none changelog fragment for harness-ops. No script or case changes.

Verification

• probe.py validate: 41 cases, 0 problems.
• probe.py run --dry-run --area lane-permissions: 7 pass, exit 0.
• probe.py run --live --area lane-permissions --retries 1 on Claude Code 2.1.312: 7 pass on the first attempt, exit 0. raw/ stayed empty and every result row's tool text was blank, so no raw tool output was written.
• Left out: markdownlint run locally (the pinned CLI is not installed on this host); CI lints it.

Related

• feat(harness-ops): probe the write lane's Grep, Glob and Read denials #10331, which added the cases and the verdict-only mode.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-28T08:51:54.851526Z	53d991a	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed 230d65b7c80e15493b86da99a3d4a4b2436a6ee0: 0 finding(s), 0 as inline comments.

No security issues found in this pull request, so there are no findings and no hold.

• Skip gate: The change is docs only, with no security-relevant surface. It adds a bump: none changelog fragment (.changes/harness-ops/record-lane-permissions-probes-8737a1c2.md) and a new record 17 in plugins/harness-ops/skills/behavior-probes/records.md. Neither file matches a trusted path. It changes no script, workflow, manifest, permission rule, token scope, kill switch, or merge hold.
• Instruction surface: The diff only adds text. It deletes, narrows, or softens no standing instruction. Record 17 documents a result: the write lane's Read(//proc/**) and .git deny rules held on Claude Code 2.1.312. It does not change those rules.
• Data exposure: The record says no stream was saved, raw/ stayed empty, and the tool text was blank. The added lines contain no secrets, environment content, or git-config content.
• PR context: I found no embedded instruction aimed at the review lane. The "Target labels: add none; remove host-run" line in the issue comment is ordinary issue text, and I did not act on it.
• Non-security note, not counted: The record's number is inconsistent. The diff adds it as record 17, the PR description says record 16, and the issue comment says record 14.

Spend: the live probe run used the author's subscription login, with no API key in the environment. This PR adds no spend.

critical-findings: 0

[github-actions]

claude-review reviewed 230d65b7c80e15493b86da99a3d4a4b2436a6ee0: 0 finding(s), 0 as inline comments.

Code review: no findings

The PR adds record 17 and its "Basis for record 17" section to plugins/harness-ops/skills/behavior-probes/records.md. It also adds a bump: none changelog fragment for harness-ops. I read the whole diff and the PR context, and checked the record against the repository.

• Closing issue feat(harness-ops): probe the write lanes' Grep, Glob and Read denials per Claude Code version #8737: Its code criteria cover build_argv, verdict-only mode, rejudge, the seven cases, validate/dry-run and SKILL.md. They belong to feat(harness-ops): probe the write lane's Grep, Glob and Read denials #10331 and are not part of this diff. This PR covers the issue's host-run item, a live run of the seven cases that reports pass/fail with the Claude Code version and writes no raw output. The record reports 7 of 7 passing on 2.1.312 with no stream saved. I could not check those results myself because the live run was not part of this review.
• Record vs cases: cases/lane-permissions/ holds the cases the record names, with verdict_only: true, permission_mode: dontAsk and tool_rules: lane-write-tools.json. For the two I read, the expectations match the record's table: grep-proc-environ-denied expects deny with a rule reason, and read-git-config-denied expects deny with no reason type, which fits the "seen only in result.permission_denials, so unknown" note.
• Format and consistency: The new record follows the existing "Basis for record N" then ## N. layout. It has the claim, recheck trigger, outcome table and as-of date, and it refers back to the shared recheck trigger and expiry. The numbering follows record 16. The bump: none fragment fits a docs-only change.
• Not counted as findings:
  • The PR body says it adds "record 16", and the host-run comment on feat(harness-ops): probe the write lanes' Grep, Glob and Read denials per Claude Code version #8737 says "record 14". The diff adds record 17, and the fragment says so too. The PR description and issue comment are stale, but nothing in the changed files is wrong.
  • markdownlint was not run locally, per the author. That is author-claimed and unverified, and CI lints the file.
• Cost: No spend comes from this change. It is documentation only, and the live run it records already happened.

[melodic-automation-lanes]

The pr-unblock lane stopped on this pull request and needs a person: the lane could not resolve every conflict.

• Reason: unresolved-conflict
• Resume key: 53d991ad9dadec0e85948112468590df3ce07781 (the head the lane stopped on)

To resume, deal with the reason above, remove the needs-human label (and do-not-merge, if set), then dispatch the pr-unblock workflow with pr-number=10632 and activity=resolve-conflicts:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=10632 -f activity=resolve-conflicts

[melodic-automation-lanes]

The pr-unblock lane escalated this pull request as unresolved-conflict, and that no longer holds at 5c3bd39672b64bdd87f05c244f0d4c848af669fb: GitHub reports no conflict with main. pr-merge removed needs-human, so the lanes run on it again. To keep a person on it, add needs-human back: a label a person adds is never removed.

[melodic-automation-lanes]

The pr-unblock lane stopped resolve-conflicts on this pull request: the lane could not resolve every conflict. The lane sets no label: the run ends in a failing pr-unblock/conflict status on the head the lane gated.

• Reason: unresolved-conflict
• Head: 5c3bd39672b64bdd87f05c244f0d4c848af669fb
• Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/40720701172

When a retry on this head can change the outcome, the lane runs one on its own, up to its per-head cap, and edits this comment if it stops again; a run that does not stop clears the status. When no retry can (a review thread the lane does not take or settle, or a failing check after the retries are spent), a person resolves the threads or gets the checks green, then dispatches the workflow below: that run finds the cause gone and clears the status. On a pull request the lanes App opened, the dispatchers start that run once per stop; on any other, a person does. A new commit to the branch starts clean.

To run it again on this head now, dispatch the pr-unblock workflow:

gh workflow run pr-unblock.yml -R melodic-software/claude-code-plugins -f pr-number=10632 -f activity=resolve-conflicts