mistral-core-workflow-a
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external user content through Mistral's text-generation models. It incorporates several security guardrails to mitigate risks:
- Ingestion points: Processes user requirements and data classes within the
SKILL.mdworkflow. - Boundary markers: Explicitly instructs the agent to "keep untrusted text out of control instructions" and use "bounded prompt/data classes."
- Capability inventory: The skill is restricted to file manipulation tools (
Write,Edit) and requires explicit approval for live network requests. - Sanitization: Mandates the use of a parser and schema validator for structured output to ensure data integrity.
- [EXTERNAL_DOWNLOADS]: The skill references official API documentation and client libraries from Mistral AI. These are well-known industry resources used for technical reference regarding model capabilities and workspace limits.
Audit Metadata