The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to retrieve and decode sensitive API keys from Kubernetes secrets, exposing them in the agent's process context and logs.
Evidence: kubectl get secret supabase-secrets -o jsonpath='{.data.api-key}' | base64 -d in references/immediate-actions-by-error-type.md.
[COMMAND_EXECUTION]: The skill executes administrative commands via kubectl to modify secrets and restart deployments, and uses curl for network diagnostics.
Evidence: kubectl create secret ..., kubectl rollout restart ..., and curl -v https://api.supabase.com in references/immediate-actions-by-error-type.md.
[EXTERNAL_DOWNLOADS]: The skill queries external service status pages and application health endpoints, including Supabase's official API and status page.
Evidence: curl -sf https://api.yourapp.com/health in references/examples.md.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes unvalidated external data from curl commands which could contain malicious instructions.
Ingestion points: Output from curl health checks and status pages in SKILL.md and references/examples.md.
Boundary markers: Absent; instructions do not provide delimiters or warnings to ignore instructions in tool output.
Capability inventory: High-privilege kubectl command execution and secret management in references/immediate-actions-by-error-type.md.
Sanitization: Absent; the skill does not specify validation or filtering of remote data before processing.

supabase-incident-runbook