The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to reproduce issues and run diagnostic probes. Safety is managed through a 'Read-first default' policy and a 'Write-probe gate' that requires user confirmation for any operation that might change the system state. Additionally, a specific Deny list explicitly prohibits the use of destructive commands like rm, drop, delete, and truncate.
[DATA_EXFILTRATION]: While the skill permits network operations via curl for API debugging, it enforces mandatory Redaction Rules. These rules require that sensitive information, such as API keys, tokens, and passwords, be replaced with [REDACTED] in all console outputs and exported report files.
[INDIRECT_PROMPT_INJECTION]: The skill processes external inputs such as bug reports and error logs. The risk of these inputs containing malicious instructions that could influence tool execution is mitigated by the structured Probe Protocol, which requires the agent to verify hypotheses with execution evidence and obtain human approval for non-read-only actions.

debug