The Agent Skills Directory

[SAFE]: The skill establishes persona-based modes (Brainstorm, Implement, Debug, Review, Teach, Ship) that govern the agent's interaction style. These instructions are purely behavioral and do not attempt to bypass safety filters or override core agent instructions.
[SAFE]: The skill explicitly promotes security best practices. For example, the 'SHIP Mode' includes a mandatory check for exposed secrets and input validation before deployment.
[SAFE]: The environment is restricted via the YAML frontmatter 'allowed-tools' field to basic file discovery and reading (Read, Glob, Grep). No tools for network access, file modification, or command execution are permitted, which mitigates the risk of data exfiltration or system tampering.
[INDIRECT_PROMPT_INJECTION]: The skill defines modes for processing external data (e.g., code reviews and debugging). This constitutes an attack surface for indirect prompt injection if the processed data contains malicious instructions. However, because the skill's capabilities are restricted to read-only file operations and it does not use high-risk tools (like shell execution or network requests), the risk associated with this surface is negligible. Evidence: Ingestion occurs during Review and Debug tasks; boundary markers are not explicitly defined in the templates; capabilities are limited to file system reads; no explicit sanitization is described.

behavioral-modes