Trigger-tree tracks which docs your AI coding agent actually reads
trigger-tree is a local, zero-token tool that reveals which docs Claude Code or Codex agents actually read — and which ones they miss.
trigger-tree is a local-only tool that measures which project documentation files — CLAUDE.md, architecture notes, conventions — AI coding agents like Claude Code and Codex actually discover during a task, and which ones they never find. It works via lightweight hooks (SessionStart, UserPromptSubmit, PostToolUse) that log reads to a local JSONL file, consuming zero model tokens and sending no data to the cloud.
Rather than flagging unread files for deletion, trigger-tree marks them as review candidates, while always-loaded context and heavily-linked docs are called out as likely-keep. Its heat/cold maps, live ASCII dashboard, and evidence-backed suggestions (e.g. 'this file has 0 reads in 3 weeks, add it to docs/README.md') let teams see whether documentation edits actually reduced the agent's 'hunting ratio' over time.
Where observability platforms like Langfuse, Arize, and W&B Weave track tokens and traces, none measure which project docs get read per task. trigger-tree fills that gap, turning documentation from a hopeful artifact into monitored infrastructure with a trackable A-F health grade.