policy-starter
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is designed for a specific legal-technical workflow, generating policy drafts based on local configuration. No malicious command execution, exfiltration, or obfuscation patterns were detected.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from a local configuration file, which creates a potential surface for indirect prompt injection if that file's content is untrusted. However, the skill lacks the capabilities to perform dangerous actions based on injected instructions.
- Ingestion points: Reads configuration from
~/.claude/plugins/config/claude-for-legal-zh/ai-governance-legal/CLAUDE.md. - Boundary markers: None identified in the prompt logic.
- Capability inventory: None. The skill only outputs text and does not utilize file-system write, network, or subprocess execution tools.
- Sanitization: No sanitization or escaping of the ingested configuration data is described before it is used to populate templates.
Audit Metadata