legal-compliance

Router skill. Resolve topic, then Read EXACTLY ONE playbook from routes:. Never load the whole cluster — one topic = one file. [DOC]

When to use

• Reviewing a contract/clause for risk, obligations, or red flags → contract-review. [DOC]
• Measuring posture against a named regulation/standard → compliance-assessment. [DOC]
• Designing controls, policies, or a control library from scratch → compliance-framework. [DOC]
• NOT for: drafting net-new legal text, litigation strategy, or jurisdiction-specific advice — flag and defer to counsel. [INFERENCIA]

Routing

1. Infer topic from the request; if two fit, ask one disambiguating question — never guess. [DOC]
2. depth=deep → apply the playbook exhaustively, verify at each step; quick → essentials only. [CONFIG]
3. Read only the routed file. If the request spans topics, run them sequentially, one playbook each. [INFERENCIA]

Spine: Discover → Analyze → Execute → Validate.

Validation gate (before "done")

• Output names the regulation/contract scope explicitly; no silent assumptions. [DOC]
• Every non-obvious claim carries one Alfa-core tag ([CÓDIGO] [CONFIG] [DOC] [INFERENCIA] [SUPUESTO]), one spelling throughout. See references/verification-tags.md. [DOC]
• Each [SUPUESTO] is paired with a concrete verification step; missing critical input → stop and ask. [DOC]
• Quality gates: constitution v6.0.0 (enforcement), script-first rule. [CONFIG]
• Check output against the binary per-lane rubric in assets/ (see assets/quality-rubric.json). [CONFIG]

Anti-patterns

• Loading multiple playbooks "for context" — defeats the router; pick one. [INFERENCIA]
• Asserting compliance/PASS as fact without an evidence tag. [SUPUESTO]
• Inventing clause numbers, fines, or regulatory citations — degrade to [SUPUESTO] and flag. [DOC]