[dmccoystephenson]

Summary

Adds a universal pre-push step to Phase 4 of the generated dev-loop template: before pushing, run gh issue view <N> for every Closes #N to confirm the issue title/body matches the PR's actual work. Numbers carried forward from earlier session context or summarized cycles are a known source of wrong-issue auto-closes — connected to RESEARCH.md §4 (context rot in long-horizon runs).

The pattern was first encoded in example-a-dev-loop; this promotion makes it universal across all newly generated dev-loop skills.

Closes #23

Test plan

• No new {{placeholders}} introduced; substitution table unchanged
• Edit is prose-only (no fenced code block), so no triple-backtick escape concern
• Phase numbers in template unchanged (still 1–10); README "What it does" count unchanged
• Confirmed the new step renders cleanly in context (read lines 188–200 after the edit)

Retrofit PRs against example-l-dev-loop, example-q-dev-loop, example-a-dev-loop (canonical reword), and example-h-dev-loop are tracked separately per the promotion-with-retrofit checklist in CLAUDE.md.

---

drafted by Claude on behalf of Daniel Stephenson

[dmccoystephenson]

Self-review rubric:

• Scope: PASS — single 2-line addition to create-dev-loop.md Phase 4; no unrelated changes (see diff).
• Tests-new: PASS — the new step is exercised by gh issue view 23 run during this very cycle to confirm issue body matches PR scope; no new placeholder, so no substitution-table test needed.
• Tests-fix: N/A — not a bug fix.
• Sibling structure: PASS — Phase 4 already opens with prose before its bash block (cf. lines 209+ rubric intro); the new step matches that pattern.
• Sibling renames: N/A — no identifiers renamed.
• Docs: PASS — Phase 7 doc sources-of-truth table (create-dev-loop.md, README.md) reviewed: README describes only what /create-dev-loop produces, not the inside of the generated template, so no README change needed.
• Issue resolution: PASS — issue #23 AC was 'Phase 4 has an explicit Closes #N verification step' + 'Step 4 substitution table unchanged'; both met.
• Validation: PASS — manual validation (no unresolved {{placeholders}} introduced; no fenced code block in the addition; phase numbering unchanged) ran inline before push.
• Placeholder-consistency: PASS — grep -oP '\{\{[A-Z_]+\}\}' count unchanged after the edit (verified during validation).
• Escaping: PASS — the added step is prose-only; no triple-backtick fenced block introduced inside the template.

All items PASS; one adversarial probe (could the step's position be inside 'PR body must include' instead?) was considered but rejected — the issue spec lists both placements as acceptable and 'before pushing' reads more naturally at the top of the phase.