fix(claude): retain scratch until background cleanup completes

[zzwong]

Summary

• Keep invocation scratch until the existing Claude background job cleanup returns by removing the premature cleanup call. The final cleanup call remains in place.
• Add four synthetic-helper cases covering success, task failure, failed stop, and refused removal. At scratch cleanup, check the recorded commands, exact warnings, and job-directory effects, while preserving the original task error or successful response.
• Use TempDir-rooted file operations in the regression without adding process discovery or signaling.

This is a narrow cleanup-order fix. It does not establish ownership of detached tool processes or terminate orphaned descendants. #613 remains open.

Validation

Go 1.26.8 on Linux with keyring_nopassage, using isolated synthetic helpers:

• All four new cases passed normally and with the race detector; no race diagnostics.
• Selected existing background launch, resume, stale-job GC, and cleanup-warning regressions passed on both unchanged main and the candidate.
• The same four new cases failed on unchanged production ordering at the intended callback-time removal/warning assertions.
• Separate negative controls for disabled stop-failure injection, disabled rm-failure injection, suppressed cleanup warnings, and joined cleanup errors failed only their expected cases. These were assertion failures, not compilation or setup failures.
• All nine planned phases completed without retries, skipped selected tests, resource stops, or source changes.
• Diagnostic gofmt -l on the exact two changed Go files returned no paths; no formatting edits were needed.

No real Claude or other provider calls were made.

Hosted CI

CI run 37738010272 completed successfully at exact published head 3cb2f1e0d3b63909b4d2ca6e2cb3b5b5a59b3c6f: all 10 jobs passed, including the Linux/macOS/Windows builds, required build aggregate, test, opt-out test, static smoke, lint, identity, and title checks. This hosted result is separate from the focused local validation above and does not validate combined #660/#627 behavior or orphan-process ownership.

Scope and Integration

The published source is based on 999e59ddfc74a67ffeef3c4b4a78a8528365afe6, with exact tested Git tree 2a41227d765c5c82b7c3d77b7456919af38ebd60. Only the production cleanup-call deletion and the new regression file are included.

Locally, full adapter-package tests, full repository tests, lint, make check, standalone builds, native Darwin/Windows checks, cancellation/fallback coverage, and combined integration with #660 were not run for this candidate. This remains a qualified draft.

#660 extracts this cleanup tail into finishClaudeBG. When integrating the two changes, apply this deletion inside finishClaudeBG and preserve #660's timeout fixture rather than replacing that file with this branch's older version. Revalidate the combined source. #627 also changes other parts of subprocess.go; combined behavior is unverified.

Related Issues

Refs #613.

[monit-reviewer]

Automated PR Review

Reviewed commit: 3cb2f1e0d3b6
Profile: open-cli-collective-review - Posting as: monit-reviewer

Summary

Reviewer	Findings
go:implementation-tests	0

Reviewer Coverage

Complete coverage details are retained in the local coverage.json artifact.

• go:implementation-tests — complete (broad); scope: 2 files; examples: internal/llmadapters/subprocess.go, internal/llmadapters/subprocess_cleanup_order_test.go; body-inspected: 2 files; examples: internal/llmadapters/subprocess.go, internal/llmadapters/subprocess_cleanup_order_test.go; skipped: none; context (not coverage): 2 files; examples: docs/development.md, internal/llmadapters/subprocess_test.go; constraints: Read-only review; tests and race checks were not executed. Assessment is based on the pinned diff, cleanup implementation, and synthetic-helper assertions.

0 PR discussion threads considered. 0 summarized; 0 resolved.

---

Completed in 37s | $0.03 | openai-codex/gpt-6.1-sol | cr 0.10.328

Field	Value
Model	openai-codex/gpt-6.1-sol
Reviewers	go:implementation-tests
Engine	pi_rpc · openai-codex/gpt-6.1-sol
Reviewed by	cr · monit-reviewer
Duration	37s wall · 28s compute
Cost	$0.03
Tokens	11.4k in / 281 out

Per-workstream usage

• orchestrator-selection — openai-codex/gpt-6.1-sol
  • In: 5.7k
  • Out: 109
  • Cache read: 0
  • Cache create: 0
  • Cost: $0.01
  • Duration: 4s
• go:implementation-tests — openai-codex/gpt-6.1-sol
  • In: 2.0k
  • Out: 125
  • Cache read: 22.3k
  • Cache create: 0
  • Cost: $0.01
  • Duration: 21s
• orchestrator-rollup — openai-codex/gpt-6.1-sol
  • In: 3.7k
  • Out: 47
  • Cache read: 0
  • Cache create: 0
  • Cost: $0.01
  • Duration: 2s

[monit-reviewer]

Automated PR Review

Reviewed commit: 6548c857ead6
Profile: open-cli-collective-review - Posting as: monit-reviewer

Summary

Reviewer	Findings
go:implementation-tests	0

Reviewer Coverage

Complete coverage details are retained in the local coverage.json artifact.

• go:implementation-tests — complete (constrained); scope: 2 files; examples: internal/llmadapters/subprocess.go, internal/llmadapters/subprocess_cleanup_order_test.go; body-inspected: 2 files; examples: internal/llmadapters/subprocess.go, internal/llmadapters/subprocess_cleanup_order_test.go; skipped: none; context (not coverage): 1 file; examples: internal/llmadapters/subprocess_test.go; constraints: Review used static inspection only; tests were not executed.

0 PR discussion threads considered. 0 summarized; 0 resolved.

---

Completed in 29s | $0.01 | openai-codex/gpt-6.1-sol | cr 0.10.328

Field	Value
Model	openai-codex/gpt-6.1-sol
Reviewers	go:implementation-tests
Engine	pi_rpc · openai-codex/gpt-6.1-sol
Reviewed by	cr · monit-reviewer
Duration	29s wall · 20s compute
Cost	$0.01
Tokens	5.2k in / 132 out

Per-workstream usage

• go:implementation-tests — openai-codex/gpt-6.1-sol
  • In: 1.6k
  • Out: 81
  • Cache read: 25.3k
  • Cache create: 0
  • Cost: $0.01
  • Duration: 17s
• orchestrator-rollup — openai-codex/gpt-6.1-sol
  • In: 3.6k
  • Out: 51
  • Cache read: 0
  • Cache create: 0
  • Cost: $0.01
  • Duration: 2s