The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): No instructions were found that attempt to bypass AI safety guardrails, override system prompts, or elicit restricted behavior. The persona is strictly professional and task-oriented.
[CREDENTIALS_UNSAFE] (SAFE): While CONTEXT.md contains environment variable keys like JWT_SECRET and MYSQL_PASSWORD, they use safe placeholder values such as 'your_jwt_secret' and 'secure'. This follows safe documentation practices.
[DATA_EXFILTRATION] (SAFE): No patterns of sensitive file access (e.g., SSH keys, AWS credentials) combined with external network requests were detected. The skill explicitly forbids printing sensitive data in logs.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill identifies an attack surface where it processes 'Task Cards' from external sources (Product Planner/Reviewer).
Ingestion points: Processes task cards and coordination requests in FLOW.md and README.md.
Boundary markers: No specific delimiters or 'ignore' instructions for the content of the task cards are defined.
Capability inventory: Has the capability to perform database migrations, write source code, and execute local scripts.
Sanitization: The skill mandates the use of Joi/Zod for input validation and Knex parameterization for database queries, which provides defense-in-depth against data-level injections.
[COMMAND_EXECUTION] (SAFE): Mentions of scripts (e.g., scripts/*.sh) are restricted to local development tasks like starting the app or running tests, which is standard for this role.

backend-dev