init-project-orchestration

# Initialize project orchestration Create the project source once under `.ai/orchestration/`. Treat Claude Code and Codex files as generated adapters, never as parallel authoring surfaces. ## Resulting structure ```text .ai/orchestration/ # canonical, client-neutral source ├── manifest.yaml ├── agents/<role>.md └── commands/<workflow>.md .claude/agents/project-<role>.md # generated Claude agent .claude/commands/project/<workflow>.md .codex/agents/project-<role>.toml # generated Codex agent .agents/skills/project-<workflow>/SKILL.md .planning/roadmap.md # shared project state .planning/state.md ``` Claude exposes a workflow as `/project:<workflow>`. Codex exposes the same workflow as `$project-<workflow>`. Both adapters derive from the same neutral command body. The namespace prevents collisions with global skills. ## Phase 1: Assess 1. Resolve the project root from the argument or current directory. Confirm it is the intended research project before searching elsewhere. 2. Read the active client's project guidance. Read neutral project facts from `AI.md`, `README.md`, `docs/`, and `.context/`; do not treat the sibling client's guidance file as an additional instruction set. 3. Read `.context/field-calibration.md`, project memory, and `.planning/` when present. 4. Classify the project: - experimental: data preparation and empirical estimation; - computational: software, algorithms, simulations, or experiments; - theo...