vercel-env-puller

Warn

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess.run to execute external CLI tools (vercel) and local Python scripts (scripts/refresh_env_manifest.py) for managing environment variable manifests.
  • [EXTERNAL_DOWNLOADS]: Fetches sensitive project configurations from Vercel's cloud infrastructure using the vercel env pull command to populate the sandbox with required secrets.
  • [DATA_EXFILTRATION]: The skill handles highly sensitive production credentials, including API keys for Anthropic and Stripe, and database connection strings. While it implements defensive patterns such as ephemeral storage and log-stripping, the processing of these secrets in memory constitutes an inherent exposure surface.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external .env files which could theoretically contain malicious content designed to influence the receiving agent. Evidence chain: (1) Ingestion point: Vercel-generated .env files via vercel env pull; (2) Capability inventory: subprocess.run and file system access in SKILL.md; (3) Boundary markers: Explicit isolation within the Sandcastle sandbox and use of non-overlapping environment rules; (4) Sanitization: Basic parsing logic with no explicit content validation of variable values.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 16, 2026, 10:22 PM
Security Audit — agent-trust-hub — vercel-env-puller