The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for installing and executing code from arbitrary GitHub repositories via the skills add command. While it encourages verifying stars and install counts, it empowers the agent to integrate external, potentially unvetted code into the environment. The use of the -y flag to skip interactive confirmation further automates this risky process.
[COMMAND_EXECUTION]: The skill utilizes shell commands (skills find, skills add, nix run) to discover and modify the agent's available tools. This includes the ability to perform global installations (-g flag), which impacts the underlying system environment.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading content from external sources including GitHub and skills.sh. It identifies well-known sources such as Vercel Labs and Anthropics as examples of high-reputation providers.
[INDIRECT_PROMPT_INJECTION]: The skill parses and acts upon data from external sources (search results and package metadata) which could be manipulated to influence agent behavior.
Ingestion points: Untrusted data enters the context through the output of the skills find command and the skills.sh leaderboard.
Boundary markers: Absent. The skill does not provide delimiters or instructions to the agent to treat search results as untrusted data.
Capability inventory: The skill can execute shell commands (skills add, nix run) and perform global installs.
Sanitization: Absent. The skill relies on the agent's manual heuristic evaluation of reputation (stars/installs) rather than programmatic validation or sandboxing.

find-skills