Firehose for 24 October 2026
Agent tooling covers a Kaizen plugin for steady task breakdown and a hook that spots unrelated tasks and moves them to a fresh session. Speech recognition work includes a model that adds paragraph breaks to flat transcripts, an ASR cleanup model, and training data for both.
GitHub Repositories
3 repos created
- Kaizen Please GitHub · Agent plugin Kaizen for any project, not just code: small steady steps, task breakdown, reviews, a Kaizen assessment essay, clean uninstall. Agent skills.
- New Task Split GitHub · Tool Claude Code hook: a local ONNX classifier detects an unrelated new task and moves it to a fresh session
- Proofed GitHub · Documentation Proofed: integrated ASR + post-processing model, from speech straight to clean, proofed text.
Datasets
3 datasets created
- Agent Task Break Detector Hugging Face · Training dataset Synthetic training data for a small classifier that answers one question on every user turn of an AI agent session:
- Paragraph Dictation Passages Hugging Face · Audio dataset Data for a separate paragraph-inference model; not used to train any Hebrish model.
- Proofed ASR Cleanup Training Data Hugging Face · Audio dataset Audio paired with clean, proofed target text for training an ASR model to do its own post-processing: no fillers, repeats or false starts, numbers and money as digits, self-corrections applied,...
AI Models
2 models created
- Agent Task Break Detector ModernBERT Hugging Face · Fine-tuned model A small cross-encoder that answers one question on every user turn of an AI agent session:
- MiniLM L12 ASR Paragraph Breaks Hugging Face · Fine-tuned model Adds paragraph breaks to flat speech-to-text output. Speech recognition returns punctuated, cased text as one block; this model decides, word by word, where a new paragraph starts.