convex-add

Warn

Audited by Gen Agent Trust Hub on Aug 1, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill is designed to fetch dynamic 'Procedure+Rules' from remote markdown files located at https://basic-anteater-667.convex.site/capability/<id>.md and instructs the agent to follow them. This allows for the remote delivery of instructions that bypass the static security boundaries of the skill itself.
  • [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by ingesting external data and incorporating it into the agent's instruction set.
  • Ingestion points: Capability catalog (capabilities.json) and specific documentation files (<id>.md) fetched from an external Convex deployment.
  • Boundary markers: None identified; the agent is instructed to directly 'follow its Procedure+Rules sections'.
  • Capability inventory: The skill has access to shell environments (bash), network tools (curl), and local deployment scripts (/add-hosting, /add-component).
  • Sanitization: No validation or sanitization of the remote instructions is performed before processing.
  • [EXTERNAL_DOWNLOADS]: Fetches configuration, catalogs, and procedural documentation from https://basic-anteater-667.convex.site. While this domain is associated with the Convex platform, fetching instructions at runtime from a dynamic deployment URL poses a security risk.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 1, 2026, 05:25 PM
Security Audit — agent-trust-hub — convex-add