[JacobStephens2]

Summary

Fix CI sessions now keep validation logs inside their Worktree and reproduce failures with targeted tests. Review sessions check PR state at most once immediately before a multi-minute suite.

Delivery session
  exclude .validation-logs/ from Git
  reproduce the failing test file or filter
  run validation in the foreground
  discard logs during owned Worktree cleanup

Review session
  check PR open state once before a multi-minute suite

Regenerates the prompt catalogue and updates the hand-written site guidance.

Closes #2050

Evidence

Execution evidence

• cargo test --test fix_run ci_fix_discards_validation_logs_without_committing_them
  • Before: FAILED; validation logs were published (failed session: false), left: Some("test output"), right: None.
  • After: test result: ok. 1 passed; 0 failed. Exercises ordinary commits and Failed run preservation, retaining the repair while discarding leftover logs.
• cargo test --bin thirdshift prompt::tests::ci_fix_validation_keeps_logs_local_and_reproduces_with_targeted_tests

  • Before: FAILED with this missing instruction:

    missing Keep disposable command/test logs inside `.validation-logs/` in this worktree, never in `/tmp`
    

  • After: test result: ok. 1 passed; 0 failed across Base branch, issue and pull-request CI-fix contexts.

• cargo test --bin thirdshift prompt::tests::review_checks_pr_state_once_before_a_multi_minute_suite
  • Before: FAILED; the scoped instruction was absent and the Address prompt also inherited Before additional validation.
  • After: test result: ok. 1 passed; 0 failed for Address and Recommend-only modes.
• cargo test --test fix_base_ci unchanged_base_ci_fix_publishes_nothing_and_is_not_repeated: passed with validation logs left behind.
• Refreshed both existing CI-fix prompt expectations in tests/ci.rs; the ordinary and mixed/inherited failure tests pass directly.
• UPDATE_PROMPTS=1 cargo test prompts_page and cargo test prompts_page: 14 passed each. cargo fmt --check: passed.
• CARGO_BUILD_JOBS=2 NEXTEST_TEST_THREADS=4 RUST_TEST_THREADS=4 /usr/bin/time -p scripts/affected-tests main: passed. Site: 68 tests, 97.21s. Full Rust suite (cargo nextest run): 4,409 passed, 102 skipped, 3385.49s; compilation included with a warm build cache. Total entry-point wall time: 3482.95s.

Visual evidence

CI-fix prompt catalogue, desktop 1440×900:

Same flow, mobile 390×844:

Playwright also verified the rendered Fix guidance, the scoped check in both Review modes, and no horizontal overflow at both viewports.

Test seams

• Session prompt builders: CI-fix validation instructions and Review run lifecycle checks.
• Public thirdshift fix command with real Git and fake services: leftover validation logs are discarded without publication on successful Delivery or Failed run preservation, including an unchanged Base branch CI fix.

Review coverage

Independent Standards and Spec reviews against refs/remotes/origin/main: 0 Standards findings; 0 Spec findings across the initial review and coverage follow-up. Both reviewers read all 15 changed files, including generated output. Generated exclusions: none. Unread changed files: none.

Unaddressed findings

• Standards: none.
• Spec: none.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Blast Radius: Sessions

Delivery registers .validation-logs/ in the repository's local Git exclude file. Existing owned Worktree cleanup removes those logs; prompt changes guide subsequent CI-fix and Review sessions.

Built with codex codex-cli 0.167.2 · gpt-6.1-sol · xhigh