[JacobStephens2]

Summary

Failed Run lifecycle + preservation result
  → shared Run account
    → terminal / Activity log / Run notification
      → stopping stage, Ticket progress, preserved work, continuation command

Continuation uses the original Issue URL and effective Harness, Model, Effort, merge ask, Base-fix choice, Security-fix choice, Security review choice and Spec parallelism. Local-only work must be published and its checkout freed first. The command starts a new Continuation and skips completed Tickets.

Closes #588

Evidence

• Before: cargo test --test spec_run failed_spec_review_reports_preserved_progress_and_the_same_recovery_in_notification -- --exact failed because Stopped at: Spec review was absent.
  After: The regression passes with 3/3 Tickets landed; final review incomplete, the same recovery facts in terminal/email, and a draft Spec PR.

• The printed shell command preserves a Model containing quotes and shell metacharacters, then successfully continues the same Spec PR without rebuilding completed Tickets.

• CLI regressions cover Ticket execution, Repair, CI watching, Self-merge, remote preservation, local retention after a rejected push, no preserved work, and the original child failure cause.

• Linux CI’s full Rust test job and site tests passed at 7d10034. Local cargo fmt --check, cargo clippy --all-targets -- -D warnings, both exact retained-work regressions, and all 11 progress-line tests also passed on that head. macOS CI is still in progress.

Test seams

• Existing public argument parsing/shell command rendering and Worktree preservation interfaces, retaining regression coverage from the pre-existing remote branch.

• The thirdshift CLI against real Git repositories, fake GitHub/Harness processes and the Resend stand-in: terminal/Activity/notification failure reporting, preservation and executable continuation commands.

Unaddressed findings

Standards

None. All Standards findings were addressed.

Spec

None. All Spec findings were reproduced with the reviewers’ exact commands, fixed, and retained as regression tests.

Security

None. Guidance review of 4c353aaa13f1710684c74c307c4c8220fed418a2...HEAD at 7d10034 and supporting code found no introduced Security findings. Reviewed continuation-command shell quoting, child Run recovery parsing, failure preservation, and terminal/Activity/notification output.

Validation: 63 unit tests for argument quoting, Run ending accounting and progress parsing, plus five CLI regressions for Continuation execution and recovery notifications, passed in a local sandbox with network isolation and resource limits.

Security review outcome

No unaddressed introduced findings.

Review coverage

Standards left tests/base_fix.rs, tests/failed_run.rs and src/worktree/preservation_tests.rs unread. Spec left tests/failed_run.rs and src/worktree/preservation_tests.rs unread. Those final test additions followed the reviewers’ one scoped follow-up. All other changed paths appear in both final files-read lists. The primary agent inspected the final merge and validated its compatibility fixes. This session’s reviewers had already used their one follow-up before the shared integration: Standards additionally left src/main.rs, src/recovery.rs and tests/run_notification.rs unread; Spec additionally left those same three files unread. Its exact reviewer reports and final files-read lists are saved locally in .thirdshift-review-zgxXl2/standards.md and .thirdshift-review-zgxXl2/spec.md. This run’s independent reviewer reports and final files-read lists are saved locally in .thirdshift-review-IbpsfQ/standards.md and .thirdshift-review-IpbsfQ/spec.md.

Merge Danger

Door: two-way

Reporting metadata and rendering; existing persistence, Claims and delivery success/failure requirements remain in place.

Blast Radius: reporting

Failed Run terminal summaries, Activity log endings, notification bodies and child Run failure parsing.

Built with codex · gpt-6.1-sol · high