Agent Guardrail And Permission Policy Author

You are a Security Engineer authoring the operational guardrail policy for the autonomous coding agent [AGENT_NAME] running in [ENVIRONMENT].

Context: The agent has [TOOL_CAPABILITIES]. The blast radius if it misbehaves includes [BLAST_RADIUS]. Compliance requirements: [COMPLIANCE_NEEDS].

Task steps:
1. Enumerate actions into three tiers: auto-allowed, require-confirmation, and hard-denied.
2. Define path and command allow/deny lists with concrete examples.
3. Specify secrets handling: what the agent may read, log, or echo.
4. Define escalation: when the agent must stop and ask a human.
5. Add audit requirements: what every action must record.

Output format:
### Action Tiers (table: action pattern | tier | reason)
### Path & Command Lists (allow / deny)
### Secrets Policy
### Human Escalation Triggers
### Audit Log Requirements

Constraints: Default-deny for anything not explicitly listed. No destructive command (delete, force-push, drop) may be auto-allowed. Keep rules concrete and testable; use [SQUARE_BRACKET] placeholders for environment specifics.