chevp-workflow

A workspace-level meta-lifecycle for
cross-repository AI-assisted development

One feature often touches many repos. When that happens, single-repo processes break down.
This framework owns the space above the single repo: impact, coordination, orchestration.

MIT License · Orchestration-driven · Cross-platform (Windows, macOS, Linux)

chevp-workflow — workspace orchestration across multiple repositories

Why a separate layer?

chevp-ai-framework breaks down at the workspace boundary — this fills the gap

Concern	Per-Repo chevp-ai-framework	Workspace chevp-workflow
One repo, one lifecycle (G1/G2/G3)	✓	—
Which repos does this feature touch?	—	✓
Cross-repo commit sequencing	—	✓
Orchestrated push / PR chaining	—	✓
Primary-owner rule per plan	—	✓
Release trains across repos	—	✓

Lifecycle

3 steps, 4 quality gates (G0–G3), human approval at every transition

1

Discover

Identify affected repos, build the impact map, assign primary owner

Output: Impact map, primary-owner assignment, confirmed scope

2

Coordinate

Write the workspace plan, sequence commits, record cross-repo decisions

Output: Workspace plan, commit sequence, cross-repo ADR, stubs

3

Execute

Orchestrated commits, pushes, PRs across all affected repos

Output: Committed repos, orchestrated pushes, linked PRs, verified state

Quality Gates

G0

Entry → Discover

Workspace mode confirmed — >1 repo touched

G1

Discover → Coordinate

Impact map complete, primary owner assigned

G2

Coordinate → Execute

Workspace plan approved, commit sequence locked

G3

Execute → Done

All repos committed/pushed/verified

Roles

Six workspace-level roles operate within each step

account_tree

Workspace-SDLC

Workspace process governance, G0–G3 gates, step transitions

hub

Impact-Mapping

Which repos, which files, which dependencies between repos

dashboard_customize

Cross-Repo-Planning

Workspace plan, commit sequence, primary-owner rule

commit

Commit-Orchestration

Atomic commit sets, push order, PR chaining, cross-refs

badge

Repo-Ownership

Single owner per plan, reference stubs in secondary repos

hub_outlined

Workspace-Context

Root CLAUDE.md, when to load workspace vs. per-repo

Framework Layering

chevp-workflow sits above chevp-ai-framework and delegates per-repo work down

hub

chevp-workflow (this repo)

Workspace meta-lifecycle for features touching >1 repo. Discover → Coordinate → Execute.

Owns: impact maps, commit sequencing, orchestration, primary-owner rule.

foundation

chevp-ai-framework

Per-repo lifecycle. Context → Exploration → Production.

Owns: single-repo context, plans, ADRs, production code.

extension

domain-ai-framework

Optional domain-specific layer adding specialised rules and templates.

rocket_launch

Project Frameworks

Concrete project frameworks (e.g. nuna-ai-framework) built on top.

Delegation contract:

Workspace Plan (in primary-owner repo)
    │
    ├── Work Item A (repo X) ──▶ chevp-ai-framework: Context → Exploration → Production
    ├── Work Item B (repo Y) ──▶ chevp-ai-framework: Context → Exploration → Production
    └── Work Item C (repo Z) ──▶ chevp-ai-framework: Context → Exploration → Production

Workspace G3 passes only when every per-repo G3 has passed.

Principles

The rules that keep cross-repo work under control

Read is trivial

Workspace framework never loads for read-only intent

Write needs a plan

Any cross-repo write requires a workspace plan first

One owner per plan

Every workspace plan lives in exactly one primary repo

Stubs, not duplicates

Secondary repos reference the owner plan, never copy it

Cross-platform

All paths and commands work on Windows, macOS, and Linux

Delegates down, never up

Per-repo work delegates to chevp-ai-framework

Quick Start

Add this to your workspace root CLAUDE.md — that's it

1

Add this block to your workspace root CLAUDE.md

## Workspace Mode — when to load chevp-workflow

This directory is a multi-repo workspace (not a single project).

Read / explain / ask → stay in the sub-repo's CLAUDE.md.
Do NOT load chevp-workflow.

Any action that writes to >1 repo (edit, commit, push, PR, release)
→ load the chevp-workflow index BEFORE the first write:

    @url https://chevp.github.io/chevp-workflow/index.md

Then follow links to the files you actually need — no bundle.

2

Done

Claude loads the index via @url, then follows links to CLAUDE.md, step READMEs, templates, and guidelines on demand. Single-repo work continues to use chevp-ai-framework.

Repository Structure

Linked from index.md, loaded on demand — no compiled bundle

index.md          # Entry point — links to every file
01-discover/      # Step 1 — Impact map, primary owner
02-coordinate/    # Step 2 — Workspace plan, commit sequence
03-execute/       # Step 3 — Orchestrated commits, pushes, PRs
templates/        # Plan, impact-map, ADR, multi-commit, stub
guidelines/       # Trivial-read, primary-owner, cross-platform
integration/      # Root CLAUDE.md snippet, composition

A workspace-level meta-lifecycle forcross-repository AI-assisted development