The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to search and ingest data from external repositories and registries, creating a potential attack surface for instructions embedded in external metadata or documentation.- Ingestion points: Workflow Step 3 in SKILL.md and the Ecosystem table in references/tool-finder.md direct the agent to read content from npm, PyPI, GitHub, crates.io, and skills.sh.- Capability inventory: The skill generates software recommendations and terminal commands for package managers based on the ingested external data.- Boundary markers: The skill does not define specific delimiters to isolate external text from internal instructions.- Sanitization: Instructions explicitly direct the agent to look for 'red flags' such as 'unsafe bundled scripts', 'installer-risk wording', and 'excessive editor permissions'.- [EXTERNAL_DOWNLOADS]: The skill workflow involves identifying and recommending external software packages and installation scripts from third-party registries.- Evidence: The Ecosystem table in references/tool-finder.md lists various external registries and the search tool skills.sh for finding agent capabilities.- Safety mitigation: The instructions include a specific safeguard requiring the agent to wait for user acceptance of risk before suggesting an official installer script.- [COMMAND_EXECUTION]: The skill facilitates the generation of shell commands for installing software.- Evidence: Guidance in SKILL.md and references/tool-finder.md covers generating commands for ni, Homebrew, and other ecosystem-specific package managers.

tool-finder