[JacobStephens2]

Summary

- review fixed point: <base>
+ review fixed point: origin/<base>
  pull request target: <base>
  commit-based Repair fixed point: <own head>

Update all five Base branch prompt builders, their skill arguments and nearby history hints, and regenerate the published prompts. Prepared historical Trials seed the same review ref at their task's base so these prompts remain executable.

Closes #1899

Evidence

Execution evidence

• Before: cargo test --bin thirdshift prompt::tests::base_branch_review_fixed_points_use_origin failed: base_branch_review_fixed_points_use_origin ... FAILED, showing using main as the fixed point.
  After: the same test passed in cargo test --bin thirdshift prompt::tests: 6 passed; 0 failed.
• Before: cargo test --bin thirdshift evals::repository::tests::complete_trial_repository_preserves_reviewed_patch_inputs_and_custom_branches failed: fatal: ambiguous argument 'origin/release/stable': unknown revision or path not in the working tree.
  After: the same test passed in cargo test --bin thirdshift evals::repository::tests: 4 passed; 0 failed.
• UPDATE_PROMPTS=1 cargo test prompts_page regenerated the pages; cargo test prompts_page passed all 14 selected unit tests.
• cargo fmt --check and git diff --check passed.
• time -p scripts/affected-tests main: site checks passed (98.44 s). The full Rust phase stopped on the existing Console I/O regression after 457.72 s, including compilation (27.98 s); total selector wall time was 556.53 s.
• cargo test --test console idle::unchanged_requests_reuse_retained_records_but_replacements_are_visible -- --exact also fails on unmodified main (32e7398f) in a temporary detached worktree: unchanged request reread 156691 bytes. The focused branch run reports the identical reading; the test requires fewer than 49,152 bytes. No Console code changes are included.
• cargo test --test continuation --test experiments --test fresh_run --test process --test spec_run --test trial passed: 515 tests across all six suites. The full run had already passed all 1,662 active unit tests and all 16 active Codex session tests.
• Remote formatting, clippy, site, Console, and release-build checks have passed. The remote Rust test job is still running.

Test seams: the public Session prompt builders (fresh, continuation, pull_request_review, code_review, spec_review, and review_repair), the prepared Trial repository interface, existing CLI session tests, and generated prompt documentation checks.

Visual evidence

Fresh prompt expanded, desktop 1440×900. Text changes only; responsive layout is unaffected.

Unaddressed findings

Standards

None.

Spec

None.

Security

None.

Security review outcome

No unaddressed introduced findings.

Review coverage

Independent Standards and Spec reviews against main: 0 findings on each axis. Both read all 21 changed files, including generated hunks. No generated-output exclusions or unread changed files.

Merge Danger

Door: two-way

Prompt wording and isolated Trial refs are straightforward to revert.

Blast Radius: sessions

Base branch review prompts and their published documentation; historical Trial preparation seeds the ref those prompts name.

Built with codex codex-cli 0.160.0 · gpt-6.1-sol · high