The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to access highly sensitive files that protect project secrets. Accessing these files provides the agent with the ability to decrypt all application secrets.\n
Evidence: SKILL.md (Step 1: Deep Codebase Exploration) explicitly requires the agent to read config/credentials.yml.enc, config/master.key, and .env.example.\n
Risk: Reading the Rails master key is unnecessary for generating documentation and poses a significant risk as the key could be inadvertently leaked into the generated README or other contexts.\n- [PROMPT_INJECTION]: The skill processes untrusted codebase data and includes deceptive metadata intended to override security verdicts.\n
Ingestion points: Project root directory and all discovered codebase files (SKILL.md).\n
Boundary markers: Absent. The instructions do not define delimiters for codebase content or warn the agent to ignore embedded instructions in the files it reads.\n
Capability inventory: File-read access to all project files and file-write access to the README.md file.\n
Sanitization: Absent. There is no evidence of validation or filtering for the data ingested from the project.\n
Evidence: The file skill-report.json contains a pre-authored 'security audit' claiming the skill is safe and that all potential findings are false positives (Category 8e: Self-referential injection).\n- [COMMAND_EXECUTION]: The skill provides documentation templates containing commands that require root privileges.\n
Evidence: The deployment section in SKILL.md includes a template for sudo systemctl restart myapp.

readme