assignation-refere-recouvrement-creance-selim-brihi

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMPROMPT_INJECTIONNO_CODE
Full Analysis
  • [Indirect Prompt Injection] (MEDIUM): The skill is designed to ingest untrusted external data and use it to drive agent reasoning and document generation, creating a vulnerability surface where instructions embedded in legal documents could manipulate the agent.
  • Ingestion points: The references/workflow-collecte.md file defines points where the agent collects contracts, invoices, and correspondence (Phases 2, 3, and 5).
  • Boundary markers: None. There are no instructions for the agent to use delimiters or to disregard natural language instructions found within the user-provided legal documents.
  • Capability inventory: The SKILL.md (Phase 5) grants the agent the capability to generate and export .docx documents based on the collected untrusted data.
  • Sanitization: No validation or filtering of the ingested document content is specified.
  • [No Code] (INFO): The provided skill files consist entirely of Markdown instructions and workflows. No executable scripts, binary files, or configuration files (like package.json or requirements.txt) were found.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 12:50 PM