The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is designed to fetch and install software from external community registries, specifically clawhub and skills.sh. Content on these platforms is contributed by third parties and is not verified for security.
[COMMAND_EXECUTION]: The assistant utilizes the golembot command-line tool to manage its environment, executing shell commands to search for, list, install, and remove skills.
[REMOTE_CODE_EXECUTION]: The installation of skills from remote sources via golembot skill add constitutes a remote code execution vector. Maliciously crafted skills from these registries could execute arbitrary scripts on the host system upon installation or invocation.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface through its reliance on external data and persistent memory.
Ingestion points: Untrusted data enters the context via notes.md (read at the start of sessions) and via search results from golembot skill search.
Boundary markers: There are no instructions to use delimiters or isolation protocols when incorporating data from notes.md or search registries into the active prompt.
Capability inventory: The skill has capabilities to read/write files and execute system commands to install new tools.
Sanitization: There is no logic provided to sanitize, escape, or validate content retrieved from external sources or previous persistent memory before it is processed by the agent.

general