The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes high-privilege kubectl commands to manage a production environment.
Evidence: Includes kubectl rollout restart deployment/firecrawl-integration and kubectl apply to modify cluster state during remediation.
Evidence: Executes a local script ./scripts/firecrawl-debug-bundle.sh for evidence collection.
[CREDENTIALS_UNSAFE]: The runbook contains instructions that explicitly retrieve and decode sensitive production credentials.
Evidence: The command kubectl get secret firecrawl-secrets -o jsonpath='{.data.api-key}' | base64 -d outputs plain-text API keys into the agent's context.
[EXTERNAL_DOWNLOADS]: The skill interacts with external domains to check service health and status.
Evidence: Performs health checks via curl to https://status.firecrawl.com and https://api.firecrawl.com. These are recognized as well-known services related to the skill's purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data which could influence agent behavior.
Ingestion points: Reads cluster logs via kubectl logs and external status pages via curl.
Boundary markers: None present in the instructions to separate data from commands.
Capability inventory: High-privilege kubectl access, environment variable modification, and script execution.
Sanitization: None detected; logs and external responses are piped directly to tools like grep and jq.

firecrawl-incident-runbook