claude-feature
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements API endpoints that ingest untrusted data which is then interpolated directly into high-capability model prompts. Specifically,
app/api/ai/chat/route.tsandapp/api/ai/agent/route.tsacceptsystemprompts andmessagesdirectly from the request body without content sanitization or boundary markers. - Ingestion points:
app/api/ai/chat/route.ts(body.system, body.messages),app/api/ai/agent/route.ts(body.prompt). - Boundary markers: Absent; user-supplied content is passed directly to the Anthropic SDK.
- Capability inventory: Execution of model logic, credit deduction via
creditService, and tool execution viaexecuteTool(inagent/route.ts). - Sanitization: Input is validated structurally via Zod, but content is not filtered for malicious instructions.
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the
@anthropic-ai/sdkpackage via npm to facilitate communication with Anthropic's services.
Audit Metadata