research-initlisted

# Research init Bring one target workspace to a verified ARC setup state. Treat setup as infrastructure work, not research intent. Stop with a precise handoff to `research-onboard`, `research-brainstorm`, or `research-scope`. Read [references/setup-contract.md](references/setup-contract.md) before a protocol merge, external `RESEARCH_ROOT`, or skill-path repair. ## Boundary Own setup orchestration only: - inspect the toolbox, target workspace, research root, protocol files, skill links, interface, and Server; - install or repair repository-backed skill symlinks; - attach `AGENTS.md` and `CLAUDE.md` through managed blocks without replacing project-owned text; - initialize an empty managed root through `lib.research_state`; - validate the current managed state; - build the interface and start or reuse a healthy Dashboard Server by default; - report the resolved roots, mutations, Server URL, health, and next skill. Do not infer or commit Project, Direction, or Experiment intent. Do not create a Package, launch a Run, modify source/data/environments, write management events directly, adopt unsupported legacy roots, or commit Git changes. ## Procedure ### 1. Inspect before mutation Run the read-only classifier: ```bash python3 skills/research-init/scripts/research_init.py \ --workspace <workspace> \ inspect \ --agent codex ``` Use `--agent claude` or `--agent both` when requested. Resolve script paths through the installed `research-init` symlink in a target p