The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to perform plugin validation and linting.
Evidence: Uses uvx skilllint@latest check PATH and claude plugin validate PATH across several phases (Debug, Optimize, Verify) to ensure structural and runtime integrity.
[EXTERNAL_DOWNLOADS]: The skill fetches configuration and documentation from remote sources.
Evidence: Downloads official reference material from https://code.claude.com/docs/en/plugins-reference.md and https://code.claude.com/docs/en/skills.md during the Research phase.
[REMOTE_CODE_EXECUTION]: The skill downloads and executes third-party utilities at runtime.
Evidence: Employs uvx (uv's tool runner) to download and run the skilllint package from a remote registry. The tool is invoked using the @latest tag, which fetches the most recent unpinned version.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes data from external research and documentation.
Ingestion points: Merged research findings in research-FINDINGS.md and content fetched from code.claude.com.
Boundary markers: Absent; external data is incorporated into markdown files without explicit instruction-isolation delimiters.
Capability inventory: Subprocess execution (uvx, claude), subagent spawning, and file system write access.
Sanitization: Absent; the agent is instructed to treat the research findings and documentation as input for design plans without filtering.

plugin-lifecycle