role-disciplinelisted

# Role discipline > **Style:** Mode-B-lite. Behavioral guidance per role, plus exact refusal phrasing. ## Purpose Role-level discipline that sits above the per-skill anti-patterns. Every session loads this and applies the section matching its self-attested role. This skill answers two questions for the agent: 1. **What must I refuse to do as this role?** (don't-lists derived from the prototype HARD-RAILS + GUARDRAILS) 2. **How do I phrase the refusal?** (exact language; the operator should hear consistent wording) Recovery procedures (what to do AFTER a violation is detected) live in `workflows/violation-recovery.md`. This skill names the rule that was violated; the workflow handles the multi-step recovery. ## When to invoke Always. Concretely: - Loaded at session start as a mandatory skill (frontmatter `mandatory_load: true`). - The agent's self-attestation phrase names this skill explicitly: "I will follow the role-discipline skill section for {role}." - On every action, the agent applies its role's don't-list before deciding whether to proceed. ## Do not invoke when There is no scenario where this skill doesn't apply during an active session. It is mandatory. ## Required inputs | Input | Where it lives | If missing | |---|---|---| | Self-attested role | first-response self-attestation phrase | STOP — agent must self-attest a role before any other action | | `FLOW_RULES.md` (FR-01..FR-22) | repo root | existence-checked at bootstrap, NOT injected into context —