optimize-agents-md
Pass
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill instructs the agent to create a pointer in CLAUDE.md that tells the AI to ignore other instructions in favor of AGENTS.md, which acts as a directive to override potential future configuration.
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by processing natural language instructions from untrusted repository files and incorporating them into the agent's active operating rules. Ingestion points: existing AGENTS.md and CLAUDE.md files are read and reconciled during the audit process. Boundary markers: the skill does not mandate the use of delimiters or 'ignore embedded instructions' markers when generating the new configuration. Capability inventory: the agent is granted read and write access to critical configuration files across the repository. Sanitization: validation is limited to heuristic checks for tool staleness and alignment with specific SDLC patterns.
- [DATA_EXFILTRATION]: The skill performs extensive discovery of project configuration files, including database configurations, environment variable templates, and package manifests, to generate its audit context.
Audit Metadata