The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation instructs the agent to execute npx skills add <owner>/<repo>, which clones remote Git repositories from GitHub into the user's environment. This facilitates the acquisition of code from external and potentially unverified sources.
[REMOTE_CODE_EXECUTION]: It encourages the use of uv run for Python scripts utilizing PEP 723 inline metadata. This execution pattern triggers the dynamic resolution and installation of third-party dependencies from package registries at runtime.
[COMMAND_EXECUTION]: The skill relies on executing multiple CLI tools, including git, npx, and python3, to perform tasks such as cloning repositories, managing symlinks, and running auditing scripts like fitness.py and trigger.py.
[INDIRECT_PROMPT_INJECTION]: As a framework for writing skill instructions, it focuses on optimizing triggers and descriptions that control agent selection logic. 1. Ingestion: Processes SKILL.md content and repository structures. 2. Boundaries: Utilizes YAML frontmatter fields like compatibility and allowed-tools to define scope. 3. Capabilities: Commands include npx, git, python3, uv run, and chmod. 4. Sanitization: Recommends using skill-fitness for auditing, but lacks strict sanitization for instruction content.
[PRIVILEGE_ESCALATION]: The skill provides instructions for using chmod 0700 to modify file system permissions for directories. While aimed at securing private data, the use of permission-modifying commands is a noted capability.

skill-authoring