The Agent Skills Directory

[SAFE]: The skill consists exclusively of informational guidelines and documentation. It does not include executable scripts, external dependencies, or dynamic context injection patterns.- [INDIRECT_PROMPT_INJECTION]: The skill addresses the vulnerability surface associated with processing untrusted data in AI features. It includes a mandatory evidence chain for remediation: 1. Ingestion points: User messages and webhooks (referenced in SKILL.md Step 7); 2. Boundary markers: Present (Step 3 instructs to explicitly state data boundaries); 3. Capability inventory: Database writes (Step 4) and tool calls (Step 2); 4. Sanitization: Present (Step 2 requires validation of types, enums, and ranges for all model outputs).- [CREDENTIALS_UNSAFE]: The skill encourages secure secret management by advising against personal subscription usage and recommending that API keys be stored only in server-side tables and never returned to the client browser.

ai-features