brida-reflex-marketplace-job-fit

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted marketplace briefs and automation descriptions as part of its decision logic.
  • Ingestion points: Data extracted from the project environment and user-supplied marketplace briefs used to feed the decision logic in SKILL.md and references/custom-reflex.json.
  • Boundary markers: The instructions require normalizing only the bounded state defined by the contract and accepting only specific declared output branches.
  • Capability inventory: The skill explicitly disclaims authority for protected actions such as payments, account changes, or deployments, and lacks external file-write or network-write operations for the processed data.
  • Sanitization: Instructions mandate that state must remain non-sensitive and deterministic host code should be used for parsing.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch availability and configuration data from the vendor's hosted metadata files (e.g., /reflex/llms.txt, /reflex/agent.json). These are legitimate vendor-controlled resources used for synchronizing with live API status.
  • [SAFE]: The skill includes a security-conscious directive to 'Never place reusable credentials in Reflex state or client bundles,' promoting best practices for secret management during the implementation of the decision logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 08:22 PM
Security Audit — agent-trust-hub — brida-reflex-marketplace-job-fit