convex
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches a capability catalog and task-specific instructions from an external source (https://basic-anteater-667.convex.site/capabilities.json). This is a documented feature designed to keep the agent's procedures up to date with the Convex platform's evolving capabilities.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes procedural instructions retrieved from a remote, vendor-controlled server at runtime.\n
- Ingestion points: Procedural documentation is fetched from subpaths of https://basic-anteater-667.convex.site/ based on the requested capability.\n
- Boundary markers: The skill instructions explicitly warn the agent that served text contains procedures and to apply normal judgment.\n
- Capability inventory: The skill family possesses significant capabilities including database migration (convex-migrate), secret management (convex-env), and code repair (convex-self-heal).\n
- Sanitization: Critical operations (Tier > 0) such as those involving financial costs require explicit user confirmation.\n- [COMMAND_EXECUTION]: The skill recommends executing CLI commands such as
npx convex ai-files installto maintain local AI guideline files. These commands are part of the standard development workflow for the Convex platform.
Audit Metadata