agents-md
Pass
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes shell commands to discover project structure and configuration metadata.
- Evidence: Uses commands like
ls -la,cat package.json, andls .github/workflows/to identify the project's build, test, and linting environment. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted local files to generate instructions for AI agents.
- Ingestion points: Reads data from files such as
package.json,README.md,CONTRIBUTING.md, and other project configuration files via theReadtool. - Boundary markers: Content from these files is interpolated into the generation process without explicit delimiters or instructions to ignore embedded commands.
- Capability inventory: The skill has access to powerful tools including
Bash,Write, andEdit. - Sanitization: No sanitization or escaping of the ingested text is performed.
- Mitigation: The workflow requires a manual confirmation step ('Phase 3: Present and Confirm') where the user must review the generated content before any file write operations occur.
Audit Metadata