Stack · 2 sessions

prettier sessions with coding agents

2 sessions on prettier with a coding agent: the agent failed at least once in 1, and 0 finished without a human stepping in.

More filters →

Claude Code·feature·Data & ML·library3d

Built a 350M nanoGPT-style trainer for erotica corpus, ready for Colab/Kaggle

Takeaway: Next time I'd specify upfront that I want the split and training pipeline to preserve the conversation metadata (system/human/gpt) rather than letting the agent default to the flattened corpus.txt — that correction is what forced the whole pipeline to be rebuilt properly.

32m3interventionsPython PyTorch tokenizers0 cmt

Claude Code·docs·CLI & tooling·crocone2d

Planned three new browser-extension features by rewriting PLAN.md

Takeaway: In my first prompt I'd point directly at PLAN.md's existing sections and ask the agent to read it fully before touching the codebase, since most of the session was just exploration to understand the plan's structure and conventions.

10m0interventionsTypeScript React Vite0 cmt