folder-management

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection as it retrieves potentially untrusted data and maintains significant capabilities.
  • Ingestion points: Data enters the agent's context through folder names, descriptions, and tags retrieved from retrieval endpoints such as GET /v1/search/folders, GET /v1/folders, and GET /v1/sessions/{sessionId}/children as specified in SKILL.md.
  • Boundary markers: The skill does not provide instructions to the agent to treat metadata as data rather than instructions, nor does it define delimiters for these fields.
  • Capability inventory: The skill grants the agent the ability to delete folders (DELETE /v1/sessions/{sessionId}), modify structures (PATCH /v1/sessions/{sessionId}), and generate presigned S3 URLs for file downloads (POST /v1/sessions/files/download/urls), all of which could be subverted if malicious content in folder names is interpreted as a command.
  • Sanitization: No sanitization or escaping mechanisms are described for handling the natural language metadata received from the API responses.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 04:01 PM
Security Audit — agent-trust-hub — folder-management