The Agent Skills Directory

[SAFE]: The skill serves as a scaffolding and review utility for developing new agent capabilities. It uses a structured multi-step workflow including planning, initialization, and sub-agent review.
[COMMAND_EXECUTION]: The skill and its references include numerous examples of shell commands for file management, package installation (e.g., pip), and utility execution. These are provided as instructional examples for the agent to use when building skills and do not involve arbitrary or hidden execution.
[PROMPT_INJECTION]: The skill allows for significant user-driven content generation. There is a potential surface for indirect prompt injection if a user provides malicious requirements intended to be embedded in the generated skill files. However, the inclusion of a mandatory review step (Step 5) using a separate sub-agent acts as a procedural mitigation.
Ingestion points: User-provided examples in Step 1 and existing skill files in Step 4.
Boundary markers: None explicitly used for interpolating user input into the generated files.
Capability inventory: Uses the Agent tool in SKILL.md to spawn review processes.
Sanitization: No explicit sanitization of user-provided examples is performed prior to generation.
[EXTERNAL_DOWNLOADS]: The references/best-practices.md file contains links to official documentation at platform.claude.com, which is a trusted domain for this environment.

create-skill