blink-full-stack
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: CRITICALEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Instructions to install official vendor packages @blinkdotnew/cli and @blinkdotnew/sdk from the NPM registry. The skill also references the vendor's hosting domain blinkpowered.com for deployed application URLs.
- [CREDENTIALS_UNSAFE]: The skill instructs users in Step 7 to store their publishableKey (prefixed with blnk_pk_) in a clear-text file named AGENTS.md. While these are intended for client-side use, storing credentials in project documentation increases the risk of exposure.
- [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to create AGENTS.md and CLAUDE.md files containing project configuration and instructions. This creates a surface where project metadata influences future agent actions in the directory.
- Ingestion points: AGENTS.md and CLAUDE.md files at the project root.
- Boundary markers: Absent in the suggested markdown templates.
- Capability inventory: Access to CLI commands for database queries (blink db), backend deployment, and hosting operations.
- Sanitization: No validation or sanitization is suggested for the values written to these configuration files.
Recommendations
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata