brida-reflex-content-policy-review
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches live configuration and documentation from the author's (brida-ai) official hosted resources, including
/reflex/llms.txt,/reflex/agent.md, and/reflex/agent.json. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and classify untrusted external content. To mitigate injection risks, it provides specific instructions to the agent to stick to bounded policy categories, perform deterministic checks first, and explicitly defines an authority boundary where the output is recommendation-only and cannot trigger protected actions.
- Ingestion points: Event or state feeding the decision as specified in SKILL.md.
- Boundary markers: Explicit instructions in SKILL.md and references/custom-reflex.json to use only provided categories and not invent policy.
- Capability inventory: Limited to recommendation; SKILL.md explicitly denies permission for payments, deploys, account changes, or destructive mutations.
- Sanitization: Recommends running deterministic checks (deny/allow lists, malware signatures) prior to semantic classification.
Audit Metadata