vercel-incident-runbook
Fail
Audited by Gen Agent Trust Hub on Mar 24, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The reference file 'references/immediate-actions-by-error-type.md' contains a command that retrieves a Kubernetes secret and decodes it:
kubectl get secret vercel-secrets -o jsonpath='{.data.api-key}' | base64 -d. This action exposes sensitive API keys in plain text within the agent's output and session history. - [COMMAND_EXECUTION]: The skill uses
kubectlto perform high-privilege operations, including creating secrets, updating deployment environment variables, and triggering rollouts in production environments. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of untrusted external data.
- Ingestion points: Deployment logs and Vercel status page content (as instructed in 'SKILL.md').
- Boundary markers: No delimiters or specific warnings are provided to the agent to isolate the processed log data from instructions.
- Capability inventory: Administrative access via
kubectl(for secret and environment management) and network access viacurl. - Sanitization: No sanitization or validation of the ingested log content is performed before the agent interprets the data.
Recommendations
- AI detected serious security threats
Audit Metadata