test(browser): retain Agent Task process teardown evidence after failure

[seonghobae]

Partial implementation of #28, based on current PR #142 head 015025f2539e4fb1dbd7d259ec22dad50f944396.

Buyer / reliability boundary

After the controlled browser's PID/start-time identity is captured, a reviewed browser failure retains a bounded failure type and an observed root-process teardown result. Temporary-profile cleanup is recorded separately. Successful cleanup never converts a failed task into a pass. Raw exception messages and command paths are not retained.

This is evidence about the original controlled Chrome root identity only: descendant-process teardown, cross-platform supervision, arbitrary browser recovery and OS-wide orphan absence remain separate work.

Current review repair and RED → GREEN

Current head: 44fd9a450f864feff5cf2ba2883425a71ba10b9b; parent: 015025f2539e4fb1dbd7d259ec22dad50f944396. Runtime behavior is unchanged from c1dd380be91c0604b797b6914f8cfef2e96f99b7.

A new release-record regression failed before the missing Unreleased entry and evidence-limit documentation were added. A controlled regression invokes both real browser-pass and isolated-trial helpers without launching a browser. It checks observed exit, a surviving identity, and process-observation failure; each remains a failed trial, driver shutdown occurs, and private exception text does not enter the result.

If process observation itself fails, termination remains unproven. The outer trial records the bounded observation-error type and profile cleanup, without a termination flag; the original browser-failure type is not retained in this fallback record. This limitation is documented rather than replaced with fabricated success or an unimplemented causal chain.

Exact-current local verification

• Five focused failure/cleanup contracts pass; all 192 repository Python contracts and compileall pass.
• Rust 1.97.1 formatting, workspace/all-target check and tests, strict Clippy, and strict rustdoc pass.
• Pinned nightly coverage passes: 415/415 functions, 3555/3555 lines, 4444/4444 regions, and 476/476 branches.
• The production compatibility runner is byte-for-byte unchanged by this review repair.

Local controlled tests are not real Linux process or pinned-Chromium runtime evidence. Fresh exact-head hosted checks must be evaluated independently; queued or historical successful runs are not current acceptance.

Historical provenance — not current acceptance

The original test-only head e847d0768d7006771b6ee10d56c52e555c0fc168 produced the failure-path retention RED. An earlier stack integration 452927dd0c6040fc7814b277e553ce6eff9e1479 had successful CI 32606887326 and Manifest V3 Compatibility 32606887322; those runs do not attest this current head or base.

Stack / authority

This remains Draft while #142 and earlier prerequisites are active. #144 must adopt and revalidate this exact head before downstream acceptance. Informational thread resolution is not a formal eligible approval, and no local result substitutes for required hosted CI, current rules, or independent approval. No merge, workflow, secret, ruleset, tag, release or publication mutation is included.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[devin-ai-integration]

Devin Review found 2 potential issues.