fact-checker
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts on arbitrary user documents, exposing it to potential indirect prompt injection if an ingested document contains adversarial instructions designed to hijack the agent context.\n
- Ingestion points: Untrusted user documents are parsed and scanned for verification in Step 1 of
SKILL.md.\n - Boundary markers: No explicit data boundaries or delimiters are defined to isolate the processed document text from the core prompt instructions.\n
- Capability inventory: The skill guides the agent to perform web searches and execute file modifications via the
Edittool as shown in Step 5 ofSKILL.md.\n - Sanitization: Content filtering or escaping is absent, but the risk is minimized by a mandatory user-in-the-loop verification step requiring explicit confirmation before any file updates are executed.
Audit Metadata