The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell command execution for repository operations and test verification. Analysis of the implementation in scripts/verification_evidence.py and scripts/capture_worker_observation.py shows that commands are executed using list-based arguments without a shell, which is a secure practice that prevents shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes outputs from sub-agents and external sources like Jira. It mitigates injection risks by using strict JSON schema validation for all worker returns (e.g., scripts/validate_worker_terminal.py) and maintaining a clear authority boundary between the orchestrator and workers.
[DYNAMIC_EXECUTION]: While the skill runs user-defined verification suites, it implements a security filter in scripts/worker_contract.py that forbids shell chaining operators (like &&, ||, ;, |) and ensures all commands start with a specific trusted prefix (rtk).
[SAFE]: The skill demonstrates a strong security architecture including deterministic hashing for state integrity, file-level locking for concurrency safety, and strict path validation to prevent traversal outside of authorized roots.

krt-swarm-seneschal