The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of third-party code through the command npx skills add <owner/repo@skill>. This allows the agent to pull and execute content from arbitrary GitHub repositories.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands to manage the skill environment, including npx skills find, npx skills add, npx skills check, and npx skills update. It specifically uses the -y flag to skip user confirmation prompts during installation and the -g flag for global installation.
[EXTERNAL_DOWNLOADS]: The skill interacts with the external domain skills.sh to retrieve leaderboard data and skill information. It also references external sources such as GitHub for package retrieval.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process metadata, descriptions, and potentially instructions from external skill packages discovered via the npx skills find command or the skills.sh website. This surface could be exploited if an attacker publishes a skill with malicious instructions in its documentation.
Ingestion points: Output from the npx skills find command and data retrieved from https://skills.sh/.
Boundary markers: The skill does not define explicit boundary markers for the data it processes, though it instructs the agent to manually verify reputation before recommendation.
Capability inventory: Includes shell execution (npx), network access to download packages, and global file system modification (-g flag).
Sanitization: The skill relies on heuristic checks such as install counts, GitHub stars, and source reputation rather than programmatic sanitization of the ingested content.

find-skills