clay-incident-runbook

Warn

Audited by Gen Agent Trust Hub on Mar 25, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes kubectl commands in references/implementation-guide.md to modify production cluster resources, including updating deployment environment variables (kubectl set env) and triggering rolling restarts (kubectl rollout restart). It also executes a local script ./scripts/clay-debug-bundle.sh for evidence collection.\n- [CREDENTIALS_UNSAFE]: The implementation guide provides a command to extract, decode, and display sensitive API keys from Kubernetes secrets (kubectl get secret clay-secrets -o jsonpath='{.data.api-key}' | base64 -d), exposing them to the agent's environment.\n- [PROMPT_INJECTION]: The skill ingests untrusted input from application logs and external status endpoints, creating a potential surface for indirect prompt injection.\n
  • Ingestion points: Log data is retrieved using kubectl logs in references/implementation-guide.md.\n
  • Boundary markers: No delimiters or warnings are used to distinguish untrusted log content from instructions.\n
  • Capability inventory: The skill has significant capabilities to modify the cluster environment via kubectl as documented in references/implementation-guide.md.\n
  • Sanitization: There is no evidence of filtering or validation for the data ingested from logs or metrics.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 25, 2026, 04:41 PM