The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external sources including issues, comments, diffs, and logs. This presents a surface for indirect prompt injection attacks. The skill mitigates this by explicitly instructing the agent to treat such content as task data only, rather than authority to weaken safety filters or disclose secrets.
[COMMAND_EXECUTION]: The skill is designed to run automated tests and configured development checks. It includes instructions to use safe, disposable resources when handling untrusted code to minimize the risk of malicious execution.
[DYNAMIC_EXECUTION]: The skill generates and executes implementation code and validation evidence at runtime to verify that requirements are met, which is the primary intended function of the tool.

implement-contract