write-agents-md
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted repository files to generate agent instructions, creating a surface where malicious repository content could influence agent behavior. 1. Ingestion points: AGENTS.md files, human documentation, configuration files, task files, CI configurations, and package documentation. 2. Boundary markers: Employs directory-based scoping where the closest file takes precedence and uses the
[UNREVIEWED]prefix to tag new rules for user review. 3. Capability inventory: File system read/write for AGENTS.md management and the ability to verify project-specific commands. 4. Sanitization: Instructs the agent to exclude secrets and unverified speculation, and to confirm command validity. - [COMMAND_EXECUTION]: The skill requires the agent to 'Verify commands and paths from files that tools can run or read,' which involves executing CLI commands discovered within the repository environment to ensure their functionality.
Audit Metadata