git:20261011.ce8211e to git:20261016.9d5cb74

2 added, 2 removed. Audit A to A.

# CLAUDE.md — Quorus

## Shared Context Protocol

- **Read `CONTEXT.md` at the start of every session.** It contains the current project state, what's in progress, recent changes, architectural decisions, and next priorities. It is the shared memory between all contributors' Claude instances.

+ **Read `docs/internal/CONTEXT.md` at the start of every session.** It contains the current project state, what's in progress, recent changes, architectural decisions, and next priorities. It is the shared memory between all contributors' Claude instances.

- **Update `CONTEXT.md` after every significant change:**

+ **Update `docs/internal/CONTEXT.md` after every significant change:**

- Add your changes to "Recent Changes" (keep last 10, drop oldest)

- Update "In Progress" when starting/finishing work

- Update "Current State" if the system's capabilities changed

- Add to "Decisions Made" when architectural choices are made

- Update "Next Up" when priorities shift

---

## Project

Quorus — coordination layer for AI agent swarms. Real-time rooms, messaging, distributed locks.

**Stack:** Python 3.10+, FastAPI, asyncio, httpx, mcp (FastMCP), pytest, ruff, rich

**Run tests:** `pytest -v`

**From an agent worktree** (no venv of its own): `PYTHONPATH=.:packages/cli:packages/mcp:packages/sdk:packages/tui ~/dev/Quorus/.venv/bin/python -m pytest -q`

**Lint:** `ruff check .`

**Lint fix:** `ruff check . --fix`

---

## Architecture

```

Claude Code A -> MCP Server (stdio) -> HTTP -> [Quorus Relay (FastAPI)] -> HTTP -> MCP Server (stdio) -> Claude Code B

```

- `quorus/relay.py` — Central relay (FastAPI). Rooms, SSE fan-out, persistence, rate limiting.

- `quorus_mcp/server.py` — MCP server. 12 tools for coordination.

- `quorus/config.py` — Config loading. Priority: env vars > ~/.quorus/config.json > legacy fallback.

- `quorus_cli/cli.py` — CLI commands. `quorus` opens TUI by default.

---

## Rules

- Files under 500 lines. Split if larger.

- Async-first. Use `asyncio.to_thread` for blocking I/O.

- All external input validated before use.

- Never log secrets or tokens.

- Tests required for new features and bug fixes.

- Conventional commits, under 50 chars, imperative mood.

## graphify

This project has a graphify knowledge graph at graphify-out/.

Rules:

- Before answering architecture or codebase questions, read graphify-out/GRAPH_REPORT.md for god nodes and community structure

- If graphify-out/wiki/index.md exists, navigate it instead of reading raw files

- After modifying code files in this session, run `graphify update .` to keep the graph current (AST-only, no API cost)