The Agent Skills Directory

[SAFE]: The skill implements a training-free inference-time intervention to suppress self-reflection tokens, which is a documented technique for improving LLM efficiency.\n- [SAFE]: Technical review of scripts/nowait_processor.py confirms the use of standard logits processing via PyTorch and HuggingFace Transformers. The code does not contain any calls to dangerous functions such as eval(), exec(), or subprocess spawning.\n- [SAFE]: No network requests, sensitive file access, or credential harvesting patterns were found in the scripts or the metadata.\n- [SAFE]: The suppression keywords listed in references/keywords.md and implemented in the script are benign transition and reflection markers (e.g., 'wait', 'hmm', 'check').

nowait-reasoning-optimizer