[dmccoystephenson]

Summary

Every newly generated <slug>-dev-loop skill now starts with two HTML comments recording the template state at generation time:

<!-- template-version: {{TEMPLATE_VERSION}} -->
<!-- generated-at: {{GENERATED_AT}} -->

TEMPLATE_VERSION resolves to the short SHA of the create-dev-loop repo at generation time (git -C <path-to-create-dev-loop> rev-parse --short HEAD); GENERATED_AT resolves to an ISO-8601 UTC timestamp. Both are HTML comments so they don't render in the skill's prose but are grep-able for drift detection.

The substitution-table descriptions tell the generating agent exactly how to derive each value at Step 4 time — the issue mentioned Step 6, but Step 4 is the natural integration point (where ALL placeholders are populated). The end behavior matches the issue's intent.

Closes #22

Test plan

• Generated skill template now opens with the two HTML comments (lines 92–93 of create-dev-loop.md)
• Substitution table has new rows for TEMPLATE_VERSION and GENERATED_AT with concrete commands for capture
• grep -oP '\\{\\{[A-Z_]+\\}\\}' confirms both placeholders appear in the template AND have substitution-table rows — placeholder-consistency invariant holds
• HTML-comment form (<!-- ... -->) chosen so the comments don't render in skill output but remain grep-able
• Localization verified: # <slug>-dev-loop heading at line 92 before edit; insertion above it lands the comments at the very top of the rendered skill
• Diff 8 LOC, 2 files — well under scope ceiling (PR Add hard PR scope gates: LOC ceiling, do-not-auto-merge paths #30)
• Per the rule shipped in PR Promote 'verify each Closes #N' to template Phase 4 #26, ran gh issue view 22 and confirmed title/body match
• Per the rule shipped in PR Track implementation outcomes back into RESEARCH.md #27, added an Implementations entry under RESEARCH.md §8
• Do-not-auto-merge check (PR Add hard PR scope gates: LOC ceiling, do-not-auto-merge paths #30): modified files create-dev-loop.md, RESEARCH.md; no matches

Deferred AC

• AC Create a private GitHub repo for each new dev loop skill #4 (example-h-dev-loop Phase 9 drift-detection step) belongs in the dmccoystephenson/example-h-dev-loop repo, not this one — Phase 9 of example-h-dev-loop explicitly forbids self-modification from inside a cycle. Filing a follow-up issue against that repo is the right channel; recording it here as deferred.

---

drafted by Claude on behalf of Daniel Stephenson

[dmccoystephenson]

Self-review rubric:

• Scope: PASS — 2 files; each addition directly maps to AC 1–3 or the CLAUDE.md PR-27 rule.
• Tests-new: PASS — both new placeholders have substitution-table rows with concrete shell commands (git rev-parse --short HEAD and date -u +%Y-%m-%dT%H:%M:%SZ) that make them executable rather than aspirational.
• Tests-fix: N/A — not a bug fix.
• Sibling structure: PASS — new substitution rows match the format of neighboring rows (single-line markdown table cells with command-in-backticks).
• Sibling renames: N/A.
• Docs: PASS — Phase 7 sources (create-dev-loop.md, README.md) reviewed; README is high-level overview and doesn't mention HTML-comment headers (correctly out of scope for the README). RESEARCH.md §8 gets an Implementations entry per PR-27 rule.
• Issue resolution: PARTIAL-but-disclosed — AC 1, 2, 3 met; AC 4 (cdl-dev-loop Phase 9 drift detection) explicitly deferred in PR body because it modifies a different repo's skill, which the cdl-dev-loop skill itself forbids self-modifying.
• Validation: PASS — grep -oP '\{\{[A-Z_]+\}\}' shows both {{TEMPLATE_VERSION}} and {{GENERATED_AT}} in template body and substitution table; no fenced blocks introduced inside the outer template (HTML comments don't need escaping).
• {{EXTERNAL_SIGNAL_LABEL}}: N/A — no CI for this repo; manual validation against the diff served as the anchor.
• Placeholder-consistency: PASS — both new placeholders documented; no orphan rows.
• Escaping: PASS — HTML comments are not fenced code blocks; no triple-backtick escape concern.

Adversarial probe — 'The issue says Step 6; you put it in Step 4. Is that a deviation?' Yes, but the deviation is rationalized in the PR body: Step 4 is where ALL placeholders are populated, so adding two placeholders there is consistent with the existing architecture. The end behavior (header populated at generation time with the right SHA) matches the issue's intent. If a reviewer pushes back, an inline reference in Step 6 pointing at Step 4 is a 1-line fix.