basecite-evidence

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes customer-submitted material through extraction, OCR, and AI context generation, which creates a potential surface for indirect prompt injection if malicious instructions are embedded in the source documents.
  • Ingestion points: User-provided files uploaded and processed via the BaseCite API as described in SKILL.md.
  • Boundary markers: The workflow requires explicit confirmation of organization identifiers and caller authorization before data actions, though it lacks specific delimiters for processed text within the prompt.
  • Capability inventory: The skill performs network operations via API calls to create, track, and withdraw records, and interacts with an MCP server.
  • Sanitization: Documentation states that raw files and full text are restricted from public MCP outputs, and the skill relies on server-side status polling rather than client-side inference.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 07:45 AM
Security Audit — agent-trust-hub — basecite-evidence