adobe-policy-guardrails
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements logic in
src/adobe/guardrails/content-policy.tsto screen Firefly prompts, creating a surface where untrusted data is processed by the agent. - Ingestion points: Untrusted data enters through the
promptparameter in thescreenFireflyPromptfunction insrc/adobe/guardrails/content-policy.ts. - Boundary markers: The provided implementation does not include explicit boundary markers or instructions for the agent to ignore potentially malicious instructions embedded within the prompts.
- Capability inventory: The skill is configured with
Read,Write,Edit, andBashtools inSKILL.md. - Sanitization: The logic uses regex patterns to check for policy violations (e.g., trademarks, explicit content) but does not specifically sanitize for prompt injection techniques.
- [EXTERNAL_DOWNLOADS]: The skill references official developer documentation and resource links from Adobe's official developer domain for API policies and authentication implementation.
Audit Metadata