pattern-to-skilllisted

# pattern-to-skill ## Mission Turn repeated work into reusable skills. Draft from `pattern-observer` candidates. Review-first — never auto-activate. Per ZEREF_OS §3.5 + D4: drafts land in `skills/drafts/`. User approves via `/review-skill`. Two-Strikes Rule applies: do not draft a skill from a single occurrence. `pattern-observer` enforces this with its 3× clustering threshold. See `references/two-strikes-rule.md`. ## DRAFT (called by /review-skill or directly) ### 1. Load candidate 1. Read candidate JSON from `memory/sync/outbound/patterns/<cluster-id>.json` 2. Verify schema_version and required fields 3. If candidate already has draft in `skills/drafts/` (check by cluster_id provenance) → update existing, do not duplicate ### 2. Synthesize skill metadata - `name`: candidate `suggested_skill_name`, kebab-case, lowercase, no `zeref-` prefix - `description`: 1–2 sentences. Pattern: "<verb> <subject> based on <N> repeated events in <hours>h. Use when <trigger context>." - `trigger`: extract from member events - If verb = "wiki-write" → trigger on "user produces <subject>-style content" - If verb = "wiki-read" → trigger on "user asks for <subject> info" - Else → trigger on "user says <verb> <subject>" - `model`: default sonnet; haiku if trivial; opus if pattern shows complex multi-step work - `max_turns`: scale with cluster size — 10 for small, 25 for large - `status: draft` (explicitly marked) - `provenance`: ```yaml provenance: cluster_id: "<from candidate