[abruption]

Summary

Relates to #127; does not close it or select a macOS recovery policy.

• Record pinned-source lock lifetimes, opener/holder counterexamples, remaining evidence gaps, and the explicit choices to retain correlation, add refusal-only filtering, or restrict recovery.
• Add unpublished same-user description-sharing fixtures for fork, inheritable exec, close-on-exec and SCM_RIGHTS. Validate observations against an expected table rather than treating successful collection as a successful measurement.
• Bound and validate actor output, reject partial EOF and oversized lines, isolate cleanup failures, reserve result files exclusively, and compare start/end script hashes. Time limits report incomplete work; they do not guarantee termination.
• Separate original runs, a parser failure, the corrected worker run and the integrating session's final-source rerun. Raw results remain private; source/result hashes and scoped summaries are documented.

No runtime, eligibility, signal path, platform support, package contents, release or operations changes. The undocumented sticky flag is not positive holder evidence. Policy choices remain subject to human approval; this PR deliberately does not implement them.

Validation

• npm ci, npm run check, npm run lint, npm test (263 pass, 3 Linux-only skips), npm run smoke:package passed.
• Focused Python suite: 23 passed, no skips, ResourceWarnings treated as errors.
• Revised nominal fixture: exit 0, collected/measured/ok true, no failures, mismatches, incomplete waits or pending diagnostics. Script start/end SHA-256 matched. Environment: macOS 27.0.1 arm64, APFS, Python 3.15.0, system lsof 4.91.
• Packaging smoke confirms these research files are outside the published artifact. Staged whitespace validation passed.
• Frozen A2/B2 source review used non-interactive Claude Code 2.1.301 (response metadata: claude-opus-5-5) and Codex CLI 0.163.0 (default model; actual model not independently reported). Both were read-only. Their findings were checked against source; framing, partial EOF, cleanup, revision wording and script-identity corrections were then tested by the integrating session. Those earlier reviews do not certify the subsequent corrections; current-head repository review and CI are separate gates.

Limits

Synthetic fixtures on one same-user APFS host only. No actual Codex process, non-APFS volume, sandbox, Mach fileport or macOS 15 measurement. A remaining diagnostic can receive kernel SIGPIPE after the runner closes its pipes. Console output and raw reports may contain local paths/PIDs and must not be published unredacted.

Keep #127 open pending the policy decision and missing evidence. No merge or publication is requested by this handoff.

Current head e4795ee393b43a8a58dcd4c73eec7532908c7ecf: all 39 repository checks passed. Automatic Code Review and Security Review both completed with no findings before merge.

Merged on 2026-10-18 as 9b69ccf3957c85b00f2f3eab6da213151f34241c following user approval, without protection bypass. #122/#127 remain open for their unresolved gates and policy decisions. No release or npm publication was performed.

[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 2026-10-16T13:19:08.084041Z	e4795ee	PR opened
🔒 Security Review	✅ Completed 2026-10-16T13:17:08.489459Z	e4795ee	PR opened

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.