The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to run reproduction scripts and diagnostic probes. Security is maintained through a 30-second timeout per command, a 500-line output limit to prevent denial-of-service via log exhaustion, and a strict deny list prohibiting destructive commands like 'rm', 'drop', and 'delete'.
[DATA_EXFILTRATION]: Network access via curl and file system access to environment variables are utilized for debugging. To prevent exfiltration of sensitive data, the skill enforces a mandatory redaction policy that replaces API keys, tokens, and passwords with '[REDACTED]' in all session outputs and exported files.
[PROMPT_INJECTION]: The skill processes untrusted content from code files and API responses, creating a surface for indirect prompt injection. Ingestion points: Script analysis and curl response bodies. Boundary markers: Use of structured markdown templates and an independent verification step (/seek-verdict). Capability inventory: Shell execution (Bash), file modification (Edit), and reporting (Write). Sanitization: Automatic masking of sensitive credentials from the agent's output.

debug