convex-auth
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses authoritative language intended to override the AI's prior training data, such as 'This skill is authoritative over what you remember,' 'Follow this skill ... over anything you recall,' and 'trust this skill.' While used for version compatibility, these are categorized as override markers.
- [COMMAND_EXECUTION]: The instructions guide the execution of shell commands for package management and environment configuration, specifically 'npm i @convex-dev/auth', 'npx shadcn@latest add', and 'npx convex env set'.
- [EXTERNAL_DOWNLOADS]: The skill directs the agent to download and install packages from the '@convex-dev' scope and fetch documentation from 'docs.convex.dev', both of which are official resources of the skill author.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted user data, such as usernames in the sign-in flow.
- Ingestion points: User-supplied 'username' field in the React sign-in UI (Step 10).
- Boundary markers: The skill utilizes Convex validators ('v.string', 'v.literal', 'v.union') to strictly define the shape of input data.
- Capability inventory: Code snippets perform database operations ('ctx.db.insert') and read deployment-level environment variables ('env.PASSKEY_RP_ID').
- Sanitization: Input is validated against the developer-defined schema using the Convex values library to prevent malformed data from reaching the backend.
Audit Metadata