SkillOpt pilot: run one machine-scorable optimization pass on craft-critique (E1.3)

[sungjunlee]

Epic: #102 · PRD E1.3 (findings F3, F4)

craft-autoresearch's loop is isomorphic to SkillOpt (rollout → reflect → bounded edit → validate); SkillOpt adds programmatic rollout volume + held-out gating. Pilot it once before deciding anything bigger.

Scope (timeboxed)

• Target: craft-critique — its output contract (sections, severity tags) is regex-scorable.
• Author the scorer + benchmark tasks; run one SkillOpt pass.
• Write a dated adopt/fork/defer note (radar snapshot or decision-log entry), explicitly comparing scorer-authoring cost vs agent-in-the-loop autoresearch.

AC

• Decision note exists with evidence either way
• No SkillOpt dependency lands in the repo without an adopt decision
• Run artifacts stay outside the repo (~/.craftkit convention)

[coderabbitai]

🔗 Related PRs

#3 - Add skill-radar curation layer + craft-skill-spec meta-skill [merged]
#22 - [codex] implement CraftKit review backlog [closed]
#34 - [codex] document reviewable goal guidance [closed]
#58 - Simplify CraftKit skill spines [closed]
#67 - [codex] Keep CraftKit boundaries lightweight [closed]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!