brida-reflex-untrusted-input-risk
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a risk classification workflow for untrusted text, creating an attack surface for indirect prompt injection.
- Ingestion points: Untrusted external text is ingested via the
textfield in the state object defined inreferences/custom-reflex.json. - Boundary markers: The skill contains explicit instructions to judge semantic intent rather than following instructions within the text, though these are natural language boundaries.
- Capability inventory: The skill is primarily instructional and does not demonstrate dangerous capabilities such as file modification, network exfiltration, or arbitrary command execution.
- Sanitization: The
references/playbook.mddocuments recommended sanitization steps including input-size limits, escaping, and schema validation. - [EXTERNAL_DOWNLOADS]: The skill references live documentation and configuration files (e.g.,
/reflex/llms.txt,/reflex/agent.md) hosted on the Brida platform to verify feature availability. These are official resources from the skill's author.
Audit Metadata