The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses kubectl to perform sensitive operations such as updating environment variables (kubectl set env), creating secrets (kubectl create secret), and restarting deployments (kubectl rollout restart). It also executes a local shell script ./scripts/exa-debug-bundle.sh to collect debug data.
[DATA_EXFILTRATION]: The skill provides instructions to extract and decode sensitive API keys from Kubernetes secrets using kubectl get secret exa-secrets -o jsonpath='{.data.api-key}' | base64 -d. While intended for triage verification, this exposes raw credentials to the agent's context.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from application logs via kubectl logs. If an attacker can influence log output, they might inject instructions that the agent could follow given its high-privilege capabilities.
Ingestion points: Reads application logs from the production cluster (SKILL.md).
Boundary markers: No delimiters or instructions are present to distinguish log data from agent instructions.
Capability inventory: The skill has kubectl and curl capabilities, allowing it to modify infrastructure based on its interpretation of the data.
Sanitization: There is no evidence of log sanitization or validation before the data is processed by the agent.
[EXTERNAL_DOWNLOADS]: Fetches service status information from the official Exa status page (https://status.exa.com) and internal health endpoints.

exa-incident-runbook