handbook-updates
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection: The skill is designed to process user-supplied handbook changes and 'new language.' This represents an ingestion surface where external, potentially untrusted text is integrated into the agent's context. Without explicit sanitization or boundary markers, maliciously formatted policy text could influence the agent's auditing logic.
- Ingestion points: The workflow accepts and processes 'new language' and handbook sections in Step 1 and Step 2.
- Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands within the policy text being analyzed.
- Capability inventory: The skill possesses file read and write capabilities, specifically interacting with configuration and matter-specific files within the
~/.claude/plugins/config/directory. - Sanitization: There is no evidence of input validation or text escaping to prevent instructions within the handbook data from being interpreted as commands.
Audit Metadata