The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches skill definitions and metadata from well-known technology providers and community registries including skills.sh (Vercel Labs), aiskillstore.io, and github.com.
[COMMAND_EXECUTION]: Provides instructions for managing and installing skills using CLI tools like npx skills and curl. Security is enforced through mandatory user approval gates and requirements to inspect third-party scripts before execution.
[DATA_EXFILTRATION]: Transmits user-defined search queries to external registry APIs to identify matching skills. No sensitive local data or credentials are included in these requests.
[PROMPT_INJECTION]: The skill is exposed to indirect prompt injection because it processes SKILL.md files from untrusted remote repositories.
Ingestion points: Where untrusted data enters agent context: githubGetFileContent of remote skill files (SKILL.md).
Boundary markers: Delimiters or 'ignore' warnings: Absent; the skill relies on manual content inspection by the agent.
Capability inventory: Subprocess calls (npx, curl), file writing (references/install-reference.md), and network operations.
Sanitization: Escaping or validation: The skill instructs the agent to audit for 'Risk', 'Unsafe commands', and 'Hidden network actions' before installation.

octocode-search-skill