What problem does it solve? When running Arbor-style research workflows in Codex or Claude Code without native Arbor tools (TreeView, TreeAddNode, TreeSetMeta, RunExecutor, GitMergeBranch, report tooling), there is no deterministic way to manage idea-tree state, capture eval scores, generate executor prompts, or validate runs. This Skill provides a local helper layer that stores state in the same shape as open-source Arbor. ## Core Features & Use Cases - Local Idea-Tree State Management: Initialize, view, add, update, prune, and propagate nodes in .arbor/sessions/<run>/.coordinator/idea_tree.json with file locking and atomic writes. - Eval Score Capture and Log Parsing: Run templated eval commands, parse scores from output or cached logs (normalizing carriage-return progress logs), and record baseline/trunk metadata. - Executor Prompts, Merge Checks, and Reports: Generate executor or smoke-mode prompts, create worktrees, guard merges, validate trees with check, and produce a final REPORT.md. - Use Case: Forward-test the Arbor skill suite by copying a project to a disposable directory, initializing a short run, adding one hypothesis, generating a smoke executor prompt, recording a mocked score, and finishing with check and report. ## Quick Start Ask the agent to use the arbor-agent-tools skill to initialize an Arbor run in the current project, set the baseline score metadata, add one hypothesis node, and generate a smoke-mode executor prompt.
