The Agent Skills Directory

[SAFE]: The skill implements a reasoning workflow for plan refinement. It features a controlled loop where the agent proposes questions and gathers facts from the environment (filesystem, tools) using sub-agents. This autonomous data collection is limited by a requirement for final user confirmation before any action is taken. The skill frontmatter also includes disable-model-invocation: true, which is a restrictive configuration that limits the model's direct access to tools, and the agent policy allow_implicit_invocation: false ensures the skill is only activated via explicit user request.

batch-grill-me