Remember Colonel James Braddock? The guy who walked into a hostile jungle alone β no written plan, no kickoff meeting, no slide deck β and came out with everything solved?
Let's be honest: there's no other Braddock in the world, and there never will be. So here's a complete squad instead: defined roles, traceable artifacts, iron-clad governance, and a Kanban JSON that doesn't lie to anyone.
Braddock doesn't ask for a PRD. But you need one.
Like any special operation worth its name, every mission requires specialists:
| Agent | Role |
|---|---|
pm |
Transforms vision into PRD |
tech-lead |
Architecture, trade-offs, risks |
architect-specialist |
Structure, modules, boundaries |
backend-senior |
Implements business rules |
frontend-senior |
Interface, flows, integration |
ui-ux |
Experience, clarity, conversion |
planner |
Breaks work into epics and tasks |
qa |
Validates, flags gaps, reviews |
Braddock went in alone to solve the problem. You have an entire squad. Use it.
No mission starts in the middle. Neither does this one:
Vision β PRD β Architecture β Spec β Tasks β Implementation β Review
Each step produces an artifact. Each artifact feeds the next step.
| Skill | What it does |
|---|---|
/kickoff |
Initializes the operation, prepares the ground |
/create-prd |
PM defines the mission objective |
/create-spec |
Tech Lead + Architect + UI/UX build the plan |
/breakdown-work |
Planner breaks into epics, stories, and small bets |
/implement-task |
Executes the next eligible task |
/review-delivery |
QA + Tech Lead validate what was delivered |
your-project/
ββ .braddock/
β ββ PROMPT.md # The kickoff prompt. Paste into Claude Code.
β ββ board/
β β ββ kanban.json # The tactical map. Always up to date.
β ββ memory/
β ββ vision.md # Why the mission exists
β ββ prd.md # What needs to be done
β ββ architecture.md # How it will be done
β ββ spec.md # The technical contract
β ββ tasks.json # What each agent will do
β ββ decisions.md # Decisions that cannot be forgotten
β ββ status.md # Where we are right now
ββ .claude/
ββ CLAUDE.md # The law. Non-negotiable.
ββ settings.json # Minimal config. Braddock would approve.
ββ agents/ # The squad. One file per role.
ββ skills/ # The moves. One skill per pipeline step.
- Always read
.braddock/memory/before taking any action. - Never implement before a PRD and spec exist.
- Never assume a business rule without recording a decision.
- Maintain traceability: vision β PRD β spec β task β code.
- Before closing any step: validate consistency, flag risks, flag dependencies.
# Install in the current directory
npx braddock init
# Install in a specific directory
npx braddock init /path/to/your/projectDoes not overwrite files that already exist in the target project.
Edit .braddock/memory/vision.md with your product idea. Without this, the squad has no mission.
Paste the content of PROMPT.md into Claude Code:
You are operating inside a virtual squad system called Braddock.
Read first, in this order:
- .claude/CLAUDE.md
- ./braddock/memory/vision.md
- ./braddock/memory/status.md
Your mission is to initialize the project and prepare the squad for execution.
Rules that cannot be broken:
- Never skip steps in the pipeline
- Never implement code before an approved PRD and spec exist
- Never assume a business rule without recording it in ./braddock/memory/decisions.md
- Always maintain traceability: vision β PRD β spec β task β code
Default pipeline:
/kickoff β /create-prd β /create-spec β /breakdown-work β /implement-task β /review-delivery
Start now with /kickoff.
From there, each skill points to the next one when it finishes. The squad is self-guided.
What makes this actually work isn't having several agents with fancy names.
It's this:
- Strong governance β nobody acts outside their role
- Fixed pipeline β no shortcuts, no improvising
- Persisted artifacts β what was decided, stays recorded
- Full traceability β vision connects to PRD connects to code
Without that, it's just a bunch of AI agents yelling in the jungle.
Braddock yelled. The squad documents.
Mission started. Waiting for
/kickoff.

