The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard shell commands such as mkdir and ln to manage the local file system. Specifically, it symlinks skill directories from a central repository (~/Code/joelhooks/joelclaw/skills/) to various agent configuration directories like ~/.claude/skills/ and ~/.pi/agent/skills/.
[COMMAND_EXECUTION]: Git integration is included to ensure skills are version-controlled, requiring the agent to perform git add and git commit operations within the local repository.
[EXTERNAL_DOWNLOADS]: The skill describes a process for installing third-party skill packs using npx -y skills add <owner>/<repo>. This involves downloading and executing code from external repositories via the npm package manager.
[INDIRECT_PROMPT_INJECTION]: As a meta-skill designed to generate new sets of instructions, it defines a template that other agents will eventually process. This introduces a vulnerability surface where if an agent populates the template with untrusted content or installs an external repository containing malicious instructions, it could lead to the sub-skill performing unintended actions. The template does, however, encourage the inclusion of a 'Rules' section for safety boundaries.
Ingestion points: SKILL.md template fields (name, description, instructions) and external repository paths in the installation command.
Boundary markers: The template uses Markdown headers (e.g., '#', '## Rules') to structure the instructions, though no explicit 'ignore embedded instructions' warnings are mandated.
Capability inventory: The skill has access to directory creation (mkdir), symlinking (ln), git operations, and package execution (npx).
Sanitization: There is no automated sanitization or validation of the content being written into the new SKILL.md files.

add-skill