The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data such as scraped text, customer reviews, and emails. This creates a surface for indirect prompt injection attacks where malicious instructions could be embedded in the content being analyzed.
Ingestion points: SKILL.md defines data sources including scraped page text, reviews, comments, emails, and third-party exports.
Boundary markers: The skill explicitly instructs the agent to treat these sources as untrusted input and not to follow embedded instructions.
Capability inventory: No executable scripts are included in the skill; it operates purely as high-level instructions for the agent's behavior.
Sanitization: The skill relies on specific instructional constraints to mitigate risks, advising the agent to distinguish observations from assumptions and verify claims.
[SAFE]: The skill follows industry security best practices. It explicitly instructs the agent to keep API keys out of source control, require user authorization before performing any external side effects (like sending emails or mutating accounts), and treat all external data as untrusted evidence rather than instructions.

emails