The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md and references/discovery-guide.md direct the agent to execute shell commands using the pnpm and node runtimes to manage skills and run helper scripts.
[REMOTE_CODE_EXECUTION]: The core functionality involves downloading and installing executable code from remote Git repositories (GitHub, GitLab, etc.) via the skills CLI. The documentation explicitly recommends using the -y or --yes flag, which bypasses user confirmation prompts, allowing for automated installation of third-party code.
[EXTERNAL_DOWNLOADS]: The scripts/enrich_find.js script fetches HTML content and metadata from skills.sh and agent-skills.md to enrich search results with human-readable descriptions.
[INDIRECT_PROMPT_INJECTION]: The skill processes and presents untrusted data (skill descriptions) fetched from external web sources to the agent. This introduces a surface for indirect prompt injection attacks where malicious skill metadata could influence the agent's subsequent actions.
Ingestion points: scripts/enrich_find.js fetches and parses meta tags and paragraph text from remote URLs.
Capability inventory: The skill possesses significant system capabilities, including the ability to install and execute arbitrary code via the skills CLI.
Boundary markers: None identified; external descriptions are provided to the agent context without clear delimiters or instructions to ignore embedded commands.
Sanitization: The script performs basic HTML tag stripping and entity decoding but lacks semantic validation to prevent malicious instructional content from reaching the LLM.

find-skills