The Agent Skills Directory

[SAFE]: The skill is designed for documentation and planning. It does not perform network operations, access sensitive system files (e.g., credentials, SSH keys), or execute arbitrary code.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from design documents (design.md) and user annotations. It mitigates potential risks through a mandatory 'Challenge Gate' where subagents review the plan for architectural fit and scope creep. Additionally, it enforces human review cycles and explicit approval before any artifacts are persisted. The ingestion points are limited to the design file and conversation, and its capabilities are restricted to writing the plan.json file.
[COMMAND_EXECUTION]: The skill contains no instructions for shell command execution. It explicitly prohibits modifying implementation files or performing research/execution tasks.

spec-plan