git:20260502.86df747 to git:20260503.51b67fa

1 added, 1 removed. Audit A to A.

# CLAUDE.md — Amplihack agent context

This file is the canonical entry point read by Claude Code (and equivalent

agent runtimes) when they spawn inside an amplihack-managed workspace. It is

staged to `$AMPLIHACK_HOME/CLAUDE.md` by `amplihack install` so that the

verifier's framework-asset presence check finds it.

## What amplihack provides

Amplihack is a meta-framework for agentic development workflows. The bundle

ships the `amplifier-bundle/` tree containing:

- `recipes/` — declarative multi-step workflows executed by `recipe-runner-rs`

(notably `smart-orchestrator`, `default-workflow`, `investigation-workflow`).

- `agents/`, `skills/`, `behaviors/`, `modules/` — reusable building blocks

composed by recipes and the dev-orchestrator skill.

- `context/` — philosophy, patterns, and trust-model documents that set the

contract for how generated code, tests, and PRs are evaluated.

- - `tools/` — helper scripts (e.g. `orch_helper.py`) referenced by recipes.

+ - `tools/` — bundled shell/helper assets referenced by recipes.

## Where to start

For most tasks, invoking the `dev-orchestrator` skill (which routes through

`amplihack recipe run smart-orchestrator`) is the correct entry point. See

`amplifier-bundle/context/PHILOSOPHY.md` for the project's core invariants,

including the **install-completeness invariant**: `amplihack install` must

fail loudly whenever a required component cannot be staged.

## Pointer

The full framework documentation lives under `$AMPLIHACK_HOME/.claude/` and

`$AMPLIHACK_HOME/amplifier-bundle/context/` after install.