[JacobStephens2]

Summary

Source-editing Session
  → prefer an available structured patch/edit tool
  → use the Harness's equivalent when available
  → allow code generation and narrowly justified scripts
  → scope replacements and check the diff

One shared paragraph in src/prompt.rs covers implementation, review and Repair sessions, including permitted exploratory edits. Regenerated the Markdown prompt examples and site's generated fragment from their sources.

Closes #734

Evidence

• Before: cargo test --bin thirdshift source_editing_sessions_prefer_available_structured_tools_and_allow_scoped_scripts failed because the fresh Session prompt lacked the editing preference.
  After: The same test passes across all nine source-editing prompt variants; all three prompt tests pass, including exclusion checks for read-only sessions and Resumes that retain their original guidance.
• Representative session edit: In a small Rust fixture with two identical values.push(7) statements, contextual apply_patch changed only the intended function to push(8). Before, its assertion failed ([7] versus [8]); after, compilation and both assertions passed, confirming the unrelated function still returned [7]. The Spec reviewer independently reran the fixture. This demonstrates this session's small edit; historical Tickets keep their repository from the sub-issue graph to the Run #672/E0308 evidence remains attributed to the retrospective, with no claim of measured productivity gains.
• cargo check --all-targets, cargo fmt --check, and cargo clippy --all-targets -- -D warnings passed.
• UPDATE_PROMPTS=1 cargo test prompts_page regenerated artifacts; all 11 rendering/reference checks pass on the reconciled branch, including the other session's matrix over all 15 prompt variants.
• cargo test --test fresh_run --test ci --test mergeable passed all 52 integration tests.
• The full cargo test suite passed on ed46f7c. Reconciliation with the concurrently published branch adds only a test in src/prompts_page.rs; production code and generated artifacts are identical to that full-suite run. Both structured-edit tests, all 11 prompt-reference checks, formatting and Clippy were rerun successfully after reconciliation.
• cd site && npm test passed all four browser tests against the regenerated page.

Test seams

• Session prompt rendering through src/prompt.rs's public prompt functions, including source-editing and read-only sessions.
• Generated prompt references through the existing prompts_page artifact checks.
• Representative context-scoped source edit through the available apply_patch tool, verified by the resulting diff and fixture assertions.

Review coverage

Independent Standards and Spec reviews against main found no findings. Both axes read every changed file (17 each); changed files left unread: none. Each report and its final files-read list is retained in the session's requested review directory.

Unaddressed findings

• Standards: None.
• Spec: None.
• Security: None. Guidance review of ee154e853b9917bb9a84bb65248c0d57640ade14...d0f548dfbb1279cb4351a6aaf2dbf5c1f10f23f7 covered all 17 changed files and supporting prompt assembly, Harness argument handling and HTML/Markdown rendering, using the relevant attack-class guidance and trust-model ADRs. No introduced Security findings. In an isolated offline sandbox, cargo test --offline --locked --bin thirdshift prompts_page -- --test-threads=2 passed 11 tests and cargo test --offline --locked --bin thirdshift prompt::tests -- --test-threads=2 passed 3 tests. No source changes were needed.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Revert the paragraph and regenerate the references to restore the prior guidance.

Blast Radius: Sessions

Changes source-editing instructions across Harnesses. Adds no host dependency or installation step.

Built with codex · gpt-6.1-sol · high