The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill researches and reads third-party SKILL.md files from across the internet to provide recommendations. This creates a surface for indirect prompt injection if a researched file contains malicious instructions aimed at the agent.\n * Ingestion points: SKILL.md content is fetched and read during the research phase using webfetch and GitHub search (referenced in references/online-research.md).\n * Boundary markers: The instructions lack explicit requirements for delimiters when processing the content of these external files.\n * Capability inventory: The skill possesses capabilities for bash execution, websearch, and modifying configuration files (kasetto.yaml).\n * Sanitization: There is no documented sanitization of the external markdown content before it is parsed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill uses npx to fetch and run an external CLI tool for marketplace discovery.\n * Evidence: Usage of npx -y @lobehub/market-cli in SKILL.md and references/online-research.md.\n- [COMMAND_EXECUTION]: The workflow involves executing local discovery and management tools via the shell.\n * Evidence: Execution of brief --json, git rev-parse, kst sync, and bundled scripts such as references/resolve-skillshub.sh.

skill-scout