extract

name	extract
description	Call Opus 4.7 on cached code samples to produce a structured signature.json capturing the user's coding style.
user-invocable	true
allowed-tools	["Bash","Read","Write"]

extract — code samples → signature.json

When invoked

• After capture populates .signature-cache/samples.json
• Manually via /cogsig extract
• Triggered by inject skill if signature.json is missing or stale

What it does (Day 2 target)

1. Load .signature-cache/samples.json
2. Compose a prompt asking Opus 4.7 to extract 6 signature dimensions:
   • naming_convention — snake_case / camelCase / kebab-case / mixed; preference intensity
   • comment_density — comments per 100 lines, docstring presence, inline vs block
   • function_length — p50 / p90 / max line counts
   • error_handling — try/except style, validation pattern, bare-except tolerance
   • import_organization — grouping, ordering, aliasing preferences
   • structural_preference — flat vs nested, early-return pattern, helper extraction threshold
3. Opus 4.7 returns JSON matching signature_schema.json
4. Validate (QA skill) + persist to signature.json at repo root
5. Append to signature_history.jsonl for drift tracking

What it does NOT do

• Does not inject — that's inject's job
• Does not judge "good" or "bad" style — it describes, it doesn't prescribe

Output

• signature.json — current signature, overwrites on each run
• signature_history.jsonl — append-only drift log

Status

• SKILL.md stub (Day 1)
• signature_schema.json (Day 2)
• implementation (Day 2)