The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides instructions for utilizing high-privilege credentials and accessing infrastructure secrets. \n
Evidence: Mentions and provides code using the SUPABASE_SERVICE_ROLE_KEY in SKILL.md and references/lock-connection-diagnostics.md, which bypasses Row Level Security. \n
Evidence: references/evidence-collection-framework.md contains a script that retrieves full Kubernetes secrets via kubectl get secret.\n- [DATA_EXFILTRATION]: The diagnostic script in references/evidence-collection-framework.md aggregates sensitive information into a bundle. \n
Evidence: The script advanced-supabase-debug.sh collects unredacted secrets, system logs, and network packet captures (tcpdump). \n
Note: The script saves raw secrets to a file named secrets-redacted.yaml, which could mislead users into thinking the content is sanitized when it is not.\n- [COMMAND_EXECUTION]: The skill uses potentially dangerous system-level commands for debugging. \n
Evidence: Includes commands for tcpdump, kubectl, and journalctl in the evidence collection framework.\n- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from multiple sources. \n
Ingestion points: Data enters the context from pg_stat_statements results, database error_logs tables, and Kubernetes/system logs. \n
Boundary markers: No explicit instructions or markers are used to isolate untrusted data from the agent's control flow. \n
Capability inventory: The skill has access to shell execution, database clients, and network utilities. \n
Sanitization: The skill does not implement validation or sanitization for the collected logs and query results before processing.

supabase-advanced-troubleshooting