analyze-changes
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes uncommitted git changes, which is untrusted content that could theoretically contain malicious instructions aimed at influencing the agent's output.
- Ingestion points: Reads uncommitted git changes from the current repository branch (SKILL.md).
- Boundary markers: None explicitly defined in the prompt to separate code changes from instructions.
- Capability inventory: The skill has the ability to write to files (changelog) and suggest shell commands (git commit).
- Sanitization: The skill includes instructions to exclude special characters, quotes, and non-ASCII characters from the generated commit message, which acts as a basic form of output sanitization.
Audit Metadata