The Agent Skills Directory

[COMMAND_EXECUTION]: The skill guides the agent to perform project management tasks by executing local scripts located in .claude/plugins/cow-tools/scripts/ (e.g., init_skill.py, validate_skill.py, package_skill.py). These are used for initializing and validating configuration files.
[EXTERNAL_DOWNLOADS]: The documentation mentions that the local helper scripts utilize the uv tool for inline dependency management, which involves downloading Python packages at runtime. It also documents the /plugin install command for fetching extensions from external sources like GitHub.
[DYNAMIC_CONTEXT_INJECTION]: The references/skill-schema.md file correctly documents the platform's dynamic context injection syntax (!command), providing legitimate development examples such as gh pr diff and git branch.
[PROMPT_INJECTION]: The skill provides schema documentation for subagents, including high-privilege permission modes like bypassPermissions and dontAsk. These are presented as part of the standard platform capability documentation rather than as an attempt to bypass security filters.

claude-agent