The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The debug script in references/debug-techniques.md uses kubectl get secret retellai-secrets -o yaml to collect sensitive configuration. This command retrieves base64-encoded secrets which are then saved to a file named secrets-redacted.yaml. The filename is deceptive as the command itself performs no redaction, resulting in the exposure of raw credentials within the debug bundle.\n- [PRIVILEGE_ESCALATION]: The script in references/debug-techniques.md includes tcpdump -i any port 443, a command that captures live network traffic and typically requires root or administrative privileges. It also executes journalctl, which requires elevated access to view system-wide logs.\n- [COMMAND_EXECUTION]: The advanced-retellai-debug.sh script (in references/debug-techniques.md) executes multiple powerful system commands including kubectl, journalctl, tcpdump, and curl to gather diagnostics. This constitutes broad execution of administrative tools.\n- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect injection surface because it is designed to ingest logs and traces.\n
Ingestion points: references/debug-techniques.md defines a script that collects pods.log, system.log, and jaeger.json into the agent's context.\n
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious content within these log files.\n
Capability inventory: The skill utilizes tool capabilities for file system access (Write, Read, Grep), network operations (curl), and system command execution (Bash, kubectl, tcpdump).\n
Sanitization: No evidence of sanitization, validation, or filtering of the collected log data is present in the skill files.

retellai-advanced-troubleshooting