[JacobStephens2]

Summary

The implement session, fresh and Continuation, the Spec review and the Foreign commits Repair take their review wording from one instruction in the prompt module. Every Session prompt stays byte-for-byte unchanged.

fresh ──────────┐
continuation ───┤
spec_review ────┼─ ReviewInstruction
review_repair ──┘    ├─ address agreed findings
                    └─ Unaddressed findings

The Spec review keeps thirdshift-tdd and committing. The Repair keeps appending entries marked as coming from Foreign commits and preserving the rest of the PR body.

Evidence

• Before: Four prompt builders repeat the review wording; cargo test prompts_page passes all 8 golden/rendering tests.
  After: One shared instruction supplies that wording; cargo test --bin thirdshift prompts_page passes the same 8 tests. UPDATE_PROMPTS=1 cargo test prompts_page succeeds, and git diff --exit-code issue-486 -- site/prompts/index.html prompts confirms regeneration changes nothing.
• cargo fmt --check, cargo check --all-targets and cargo clippy --all-targets -- -D warnings pass.
• cargo test: 1344 passed, 0 failed, 1 existing ignored local latency probe.
• Seams under test: Session prompts through the existing prompts-page golden tests (prompts_page_shows_what_the_binary_sends and prompt_files_show_what_the_binary_sends); factory execution through the existing CLI integration suite.
• Independent thirdshift-code-review reviews against issue-486: Standards 0 findings; Spec 0 findings. Both reviewers read the full changed file.

Merge Danger

Door: two-way

Blast Radius: prompts

Internal refactor with unchanged instructions and generated files.

Unaddressed findings

Standards

None. The Standards review reported no findings.

Spec

None. The Spec review reported no findings.

Closes #487

Built with codex · gpt-6.1-sol · xhigh