The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from multiple sources, including GitHub URLs (SKILL.md files), video transcripts, and user-provided text dumps. It lacks explicit boundary markers or specific instructions to the agent to disregard potential directives embedded in the source content. During the 'Adapt' or 'Create' modes, a malicious source file could theoretically influence the agent to perform unintended file modifications or repository operations.\n
Ingestion points: External GitHub URLs defined in ADAPT mode, user-provided text dumps in CREATE mode, and video transcripts processed via the watch-video skill.\n
Boundary markers: None; the instructions do not require the agent to treat external content as strictly data or to ignore instructions contained within it.\n
Capability inventory: File modification via the Edit tool, local repository searching via grep, and network/repository management via git and gh CLI.\n
Sanitization: None; the content is parsed and rewritten based on its own internal logic without a sanitization layer.\n- [COMMAND_EXECUTION]: The skill provides templates for executing standard shell commands such as git clone, gh api, and grep. These are used for its core purpose of repository management, such as shallow cloning external skills to temporary directories and searching local repositories for relevant files to propagate updates.\n- [EXTERNAL_DOWNLOADS]: The skill fetches skill definitions and repository content from external URLs (e.g., GitHub, agentskills.io). These downloads target well-known services and are essential for the skill's primary function of porting external capabilities to the user's environment.

skillify