The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The shell script advanced-supabase-debug.sh found in references/evidence-collection-framework.md executes kubectl get secret supabase-secrets -o yaml. This command extracts the full content of the Kubernetes secret, including base64-encoded credentials, and saves them to a local configuration file.
[COMMAND_EXECUTION]: The skill employs high-privilege system utilities such as tcpdump for network packet sniffing and kubectl for cluster-wide resource management and secret retrieval. It also provides SQL instructions to kill database processes using pg_terminate_backend, which could be used to disrupt services.
[DATA_EXFILTRATION]: Both SKILL.md and the reference scripts guide users to aggregate sensitive data—including network captures, logs, and K8s secrets—into a 'diagnostic bundle' intended for submission to external support portals. This workflow significantly increases the risk of accidental exposure of infrastructure credentials.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted data from kubectl logs, journalctl, and Edge Function logs (ingestion points) into the agent's context. The absence of boundary markers or sanitization, combined with the agent's broad capabilities (capability inventory: psql, Bash, curl), allows malicious content embedded in logs to potentially influence the agent's subsequent actions.

supabase-advanced-troubleshooting