agent-instruction-compressor
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill provides clear and constructive instructions for refining agent skills. It emphasizes the importance of preserving safety filters, destructive-operation safeguards, and error-handling logic during the compression process.
- [INDIRECT_PROMPT_INJECTION]: The skill acts on external Markdown files (SKILL.md and references/) which could contain untrusted instructions. 1. Ingestion points: Reads and modifies SKILL.md and Markdown files in references/. 2. Boundary markers: None explicitly defined in the input structure. 3. Capability inventory: Modifies local files and executes shell commands for validation (SKILL.md). 4. Sanitization: The skill explicitly instructs the agent to identify and preserve safety constraints and destructive-operation safeguards, mitigating the risk of removing security controls.
- [COMMAND_EXECUTION]: The instructions mention running the target skill's validator or packaging workflow. While this involves executing scripts, it is a standard step for ensuring the functional correctness of the compressed output and is limited to scripts already present in the user's project environment.
Audit Metadata