# CLAUDE.md — Claude adapter for Numu Angels (thin router)

> Claude Code / Claude session entry point. This is a **thin router** with **no** project knowledge — follow [`AGENTS.md`](AGENTS.md), the primary adapter. All rules, architecture, features, and the mandatory workflow live in the **Engineering Platform**, never here.


> **Platform source of truth:** `https://dev.azure.com/numuinvestment/numu/_git/numu-engineering-platform`.
> The local folder is only a working copy — **`git pull` it before every task**, and **commit + push** any change you make inside it.
> Missing it? `git clone https://dev.azure.com/numuinvestment/numu/_git/numu-engineering-platform engineering-platform` as a sibling of this repository.

## Do this first
1. Read **[`AGENTS.md`](AGENTS.md)** and follow it exactly — locate the Engineering Platform (`../engineering-platform/`, `../numu-engineering-platform`, or `ENGINEERING_PLATFORM_PATH`), load the canonical workflow, then task-scoped knowledge.
2. Load the mandatory workflow before any task work: `` `<platform>/01-foundation/ai-workflow.md` `` — the 7-gate execution loop (intake → context loading → impact analysis → implement → verify → doc sync → report).
3. If the Engineering Platform cannot be found, apply `AGENTS.md`'s knowledge-location failure behavior — **stop and request the location; never silently fall back to this repo's legacy `docs/ai/`.**

## Claude-specific notes (mechanics only — no project knowledge)
- Use the harness's plan/todo tooling to track the `ai-workflow.md` gates; it complements, never replaces, them.
- Report outcomes honestly (Constitution §10.3): failing/skipped tests are never presented as passing.

Impact analysis, high-risk approval stops, documentation synchronization, the "No documentation impact" evidence rule, and the completion report are defined **once** in `ai-workflow.md`. Do not restate them here.