The Agent Skills Directory

[SAFE]: The skill consists of markdown instructions and guidelines for a debugging workflow. It does not contain any malicious payloads, hidden commands, or automated exfiltration logic.- [COMMAND_EXECUTION]: The skill instructs the agent to perform standard development tasks such as running local Python reproduction scripts (python reproduce_xxx.py) and using standard tools like git bisect and ast-grep for code analysis. These operations are conducted within the local environment on user-provided or project files.- [DATA_EXFILTRATION]: While the skill suggests gathering diagnostic metadata (e.g., paddle.__version__, CUDA versions, hardware info), this data is intended for local debugging reports stored in the project's .paddle-agent/ directory. No network exfiltration or remote data transmission is initiated by the skill instructions.

paddle-debug