What problem does it solve? AI agents that call external tools can execute dangerous commands, leak sensitive data, or exceed intended boundaries. This Skill provides reusable Python patterns to control which tools agents may call, filter harmful content, detect malicious prompts, and maintain accountability through audit logs. ## Core Features & Use Cases - Governance Policies: Define declarative allowlists, blocklists, content filters, and rate limits as composable Python dataclasses or YAML configuration, with most-restrictive-wins composition semantics. - Intent Classification & Tool Decorators: Detect prompt injection, data exfiltration, and privilege escalation before tool execution, and wrap any tool function with a @govern decorator enforcing policy checks. - Trust Scoring & Audit Trails: Track agent reliability with decay-based trust scores for multi-agent delegation, and record every action in an append-only audit log exportable as JSONL. - Use Case: A support agent built with PydanticAI needs to search documents and create tickets but must never execute shell commands or process content containing credentials — apply the governance policy and decorator to enforce these boundaries. ## Quick Start Add governance controls to my PydanticAI agent so it can only call approved tools, blocks prompts containing secrets, and logs every action to an audit trail.
