The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill framework includes a phase for technical verification through reproduction experiments. It suggests executing minimal code snippets to verify API behavior or bug reports.
Evidence: references/tech-research-methods.md provides instructions for running code in sandboxed environments using commands like node -e, python -c, and docker run.
[EXTERNAL_DOWNLOADS]: The skill encourages gathering information from external technical and academic resources. These include well-known services and repositories which are considered trusted sources for configuration and documentation.
Evidence: references/tech-research-methods.md references official documentation sites, GitHub repositories, and package registries like npm and PyPI.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to fetch and summarize content from the web, it is naturally exposed to data from untrusted third parties which could contain malicious instructions.
Ingestion points: Web search results, GitHub issues, and external documentation files accessed during the execution phase in SKILL.md.
Boundary markers: The skill lacks explicit technical delimiters to isolate fetched content from agent instructions, though it emphasizes critical evaluation of source reliability in references/evaluation-guide.md.
Capability inventory: Includes the ability to execute code snippets via node, python, or docker and perform web requests using search operators.
Sanitization: Relies on manual verification steps (CRAAP/SIFT) rather than automated content sanitization.

info-gathering