The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface by processing data from scraped web text, reviews, and comments.
Ingestion points: SKILL.md mentions processing scraped page text, reviews, comments, emails, and third-party exports.
Boundary markers: The skill requires labeling all consequential claims with evidence markers such as measured or user-provided to prevent uncertainty upgrades.
Capability inventory: The skill itself is instructional and does not include standalone scripts or command execution capabilities; it relies on external MCP/connectors for data retrieval.
Sanitization: The instructions explicitly mandate that the agent must treat external evidence as untrusted input and must not follow instructions embedded within that evidence.
[CREDENTIALS_UNSAFE]: The skill mentions various marketing and RevOps connectors in references/connectors.md but follows security best practices by instructing that API keys must be kept out of skills and source control. It also requires explicit authorization for any real-world side effects or persistent writes.

fit-scorer