What problem does it solve? Starting an autonomous Arbor research run requires a precise contract — metric, baseline, dev/test split, budget, and constraints — plus a clean session directory. Doing this by hand is error-prone and often skips preflight checks, causing failed or misdirected runs. ## Core Features & Use Cases - Contract Generation: Inspects the target repo's README, configs, and eval entry points to produce a five-part research contract (metric, baseline anchor, ambition, scope, hard constraints) confirmed by the user. - Session & Config Setup: Initializes .arbor/sessions/<run_name>/ layout, persists ARBOR_CONTRACT.md and research_config.yaml, and honors config precedence across plugins, profiles, project YAML, and CLI. - Preflight & Scaffolding: Mirrors arbor run preflight checks (git state, eval entry point, credentials) and scaffolds benchmark plumbing via scaffold_benchmark when a target has no runnable eval. - Use Case: Point it at a fresh ML repo, and it identifies the primary metric and eval command, locates or plans a baseline, writes the contract files, and hands a launch-ready session to the coordinator. ## Quick Start Use the arbor setup intake skill to inspect this repository, define the metric and baseline, and create the Arbor run contract and session directory.
