recipe-agentfile-hygiene
Pass
Audited by Gen Agent Trust Hub on Jul 11, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill audits and processes the content of
CLAUDE.mdandAGENTS.mdfiles across the project. These files are untrusted inputs that could contain malicious instructions designed to influence the agent's classification or remediation logic. - Ingestion points: Project files matching
**/CLAUDE.mdand**/AGENTS.md(Step 1 and Step 3). - Boundary markers: Absent; the instructions do not specify delimiters or explicit "ignore embedded instructions" warnings when processing the agentfile content.
- Capability inventory: Local script execution (
.plan/execute-script.py) and file creation (solution_outline.md). - Sanitization: Absent; no escaping or validation of the agentfile content is mentioned before the classification step.
- [COMMAND_EXECUTION]: The workflow involves executing local scripts via
python3 .plan/execute-script.pyto resolve paths and validate the solution outline. While these calls are part of the defined framework for this skill, they represent a capability for shell interaction.
Audit Metadata