The Agent Skills Directory

[SAFE]: The skill instructions define a documentation-focused workflow for creating implementation plans. It does not contain any code execution, network access, or privilege escalation patterns.
[SAFE]: Data processing is limited to reading a local specification file (spec.md) and incorporating human feedback. All generated output is subject to an explicit human approval cycle before finalization, which mitigates risks of indirect prompt injection from the source files.
[SAFE]: The skill explicitly delegates implementation and execution to separate components (e.g., spec-implement), ensuring a clear separation of concerns and preventing accidental command execution during the planning phase.
[SAFE]: Although the skill mentions loading task-tracking extensions (e.g., beads), these are used within the platform's standard capability framework for organizational purposes and do not introduce unverified remote code execution.

spec-plan