The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from external sources, including GitHub repositories and raw URLs, using git clone and fetch based on user-provided arguments.
[COMMAND_EXECUTION]: The skill executes shell commands such as git clone to a temporary directory and uses grep for data scrubbing. It also interacts with a local repository script (bin/install-agents.ts) as part of the integration process.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes untrusted markdown files from external sources.
Ingestion points: The skill reads external files (agents, skills, and commands) from repositories or URLs provided at runtime.
Boundary markers: There are no explicit instructions to the agent to treat the source data as untrusted or to ignore embedded instructions within the harvested files.
Capability inventory: The skill has the ability to write new files to the repository, execute shell commands, and perform network operations.
Sanitization: While it includes logic to scrub machine-specific strings (like home directories and IP addresses) and report secrets, it does not explicitly sanitize the source content for malicious LLM instructions.

sync-agenticat