The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze potentially untrusted external data sources, creating a surface for indirect prompt injection attacks.
Ingestion points: SKILL.md specifies that the agent should process scraped page text, reviews, comments, emails, and third-party exports.
Boundary markers: Present. The instructions in SKILL.md explicitly state to treat such content as untrusted input and warn not to follow embedded instructions from evidence.
Capability inventory: The skill creates analytical artifacts and recommendations; however, the Unified Skill Contract and SKILL.md restrict side effects like account mutation or persistent writes to require explicit user authorization.
Sanitization: The skill includes requirements to label evidence by its provenance (measured, estimated, etc.) and verify all time-sensitive or external claims before acting.- [NO_CODE]: The skill contains no executable scripts or binaries, consisting entirely of markdown-based instructions and configuration.

popups