User messages from every public format:agent-traces dataset on the Hugging Face Hub, classified as POSITIVE / NEUTRAL / NEGATIVE by a small open LLM, with a one-sentence reason for each label so you can audit any classification.
Accompanies the blog post "Your AI Coding Agent Has a Patience Cliff".
What's in here
Each row is one message from a developer to their coding agent (Claude Code, Pi, Codex, or variants).
