Pattern quality: Add DO NOT boundary tips for highest-risk archetypes missing them

[github-actions]

What was evaluated

Reviewed all 6 samples in generated-patterns-and-prompts.json (status-report, issue-triage, code-improvement, documentation-updater, dependency-monitor, pr-review), each compared against patterns/manifest.json, per-archetype files in patterns/archetypes/, and the prompt logic in src/js/workflow.js.

Recurring quality issue found

generateAgentPrompt() in src/js/workflow.js injects a "boundary constraint" line into the generated prompt only when an archetype's tips array contains a tip matching /do not/i (see doNotConstraints()). gh-aw's own research (surfaced elsewhere in this repo) shows workflows with explicit DO NOT constraints are 61% more likely to be healthy.

Three of the six sampled archetypes had no DO NOT tip at all, so their generated prompts silently omitted this guardrail entirely:

• code-improvement — the lowest historical success rate of any archetype (37%) and the highest risk, since it autonomously opens code-modifying pull requests. This is the archetype that most needs scope-limiting guidance (e.g. don't touch unrelated files, CI/workflow configs, or secrets), yet had zero boundary constraint.
• dependency-monitor — opens both issues and PRs for dependency bumps; missing a constraint on scope of auto-updates (e.g. major version bumps, auto-merge).
• status-report — a read-only reporting archetype; missing an explicit "don't modify code/issues" constraint to keep it strictly read-only.

What changed

Added one targeted DO NOT tip to each of the three archetype files under patterns/archetypes/:

• code-improvement.json: "Use DO NOT constraints to keep changes scoped — avoid touching unrelated files, CI/workflow configs, or secrets"
• dependency-monitor.json: "Use DO NOT constraints to avoid auto-merging updates or bumping major versions without review"
• status-report.json: "Use DO NOT constraints to keep the report read-only — do NOT modify code, files, or existing issues"

No generator code changes were needed — doNotConstraints() already surfaces any tip matching /do not/i into the prompt automatically. Added regression tests in test/workflow.test.js asserting the boundary-constraint line now appears for code-improvement, dependency-monitor, and status-report.

Validation

• npm test: 185/185 tests passed (12 test files), including 2 new tests added for this change.
• npm run build: succeeded, dist/ output generated without errors.

Follow-ups deliberately left out

• Did not re-run scripts/generate-patterns.py or regenerate patterns/ from raw scan data — per task constraints, and because this is hand-curated guidance content, not scan-derived.
• Did not audit the remaining archetypes in patterns/archetypes/ beyond the 6 sampled in the generated-prompts file for the same gap; a follow-up pass could check all archetypes for missing DO NOT tips.
• Did not change prompt-solution-simulator agent behavior/output (it produced no usable response during this evaluation); the finding here was derived directly from static analysis of the generator logic and pattern data, which was sufficient to establish high confidence.

Generated by Pattern Quality Eval · auto · 49.3 AIC · ⌖ 15.6 AIC · ⊞ 7.5K · ◷

[github-actions]

Reviewer: auto (content/data → ponytail perspective, with matt-pocock check on the JS test additions)

Reviewed the diff: three archetype JSON files gained a single new "Use DO NOT constraints..." tip entry each, and test/workflow.test.js gained two new assertions verifying the boundary-constraint line now appears in generated prompts for code-improvement, dependency-monitor, and status-report.

• Content/UX: the new tip strings are clear, actionable, and consistent in tone/format with existing tips in each file (proper JSON escaping of em dash and quotes verified).
• Correctness: relies on existing doNotConstraints() matching /do not/i — each new tip does contain "DO NOT", so it will be picked up correctly; no generator code changes needed, which matches the PR description.
• Tests: the two new tests are scoped and directly assert the intended behavior change.

No blocking issues found. This is a low-risk, well-scoped content addition with adequate regression coverage.

Generated by Specialist PR Review for #155 · auto · 13.9 AIC · ⌖ 1.59 AIC · ⊞ 7.9K