Open-source Claude Code plugin · v0.1

Your senior product & engineering team, in one command.

Type /raftly and an idea becomes a researched, decided, buildable production blueprint. Then Raftly keeps guiding your repository, building the next piece, verifying it against evidence, and telling you honestly whether you're ready to launch.

In Claude Code: /plugin marketplace add roszhan2684/raftly
claude · ~/tally
Why Raftly

Not a roadmap prompt. Not a code generator.

Code generators write code. Roadmap prompts write plans. Neither remembers why a decision was made, stops you overbuilding, or tells you the truth about launch readiness. Raftly is the organization around the code.

week-by-week roadmapdependency-ordered build graph

Plans that can actually be built

Capabilities in dependency order, each with a feature spec and a self-contained implementation prompt.

"looks done to me"evidence-gated status

"Done" means verified

Nothing becomes VERIFIED, COMPLETE or PASS without a recorded test, review or measurement. The scripts enforce it.

forgets every sessionpersistent .raftly/ memory

Remembers why

Decisions, overrides, assumptions, risks and lessons live in your repo, as plain Markdown and YAML you can diff.

How it works

The judgment of a senior organization, behind one command

Raftly routes every decision to the right senior role: product strategy, design, architecture, security, SRE and 17 more. It decides what it can, researches what it can't, and asks you only the genuine business questions.

Orient

Reads your repo and its .raftly/ state. Works out the mode: greenfield, build, review, recovery, migration, launch.

Research

Checks what's current and uncertain against primary sources. Anything it couldn't verify is labelled not verified, never invented.

Decide

Routine choices are made silently. Important ones become recorded recommendations. Only business calls come to you, batched into one question.

Record

Every major decision becomes an ADR with options, confidence, consequences and a revisit trigger. Your overrides keep Raftly's original advice.

Design

2–3 genuinely different UI directions (different interaction models, not different colours), with a recommendation, full screen states and accessibility.

Plan

The simplest architecture that meets the actual requirements. Feature specs, a build graph and one prompt per capability.

Build & verify

/raftly build next implements the next unblocked node, runs your tests and records real evidence before anything is marked done.

Launch honestly

/raftly launch runs a production gate fitted to your product. No evidence means no PASS. "Not applicable" is a valid answer.

Proportionate by design

It refuses to overbuild

Same command, very different products. Raftly decides which production controls apply to your profile, so a calculator is never judged on missing Kubernetes.

Tally: a calculator

/raftly I want to build a calculator mobile app and website.
Launch checks that apply
16
Build nodes
7
Decisions recorded
8
client-onlydecimal mathsofflinebackendauthdatabaseKubernetes

SalesLens: B2B AI SaaS

/raftly Build a B2B AI sales coaching platform that ingests sales calls…
Launch checks that apply
42
Build nodes
14
Decisions recorded
12
multi-tenantrow-level securityAI eval gatebillingconsentfounder decision: pricing
Build graph

Capabilities in dependency order, not a calendar

Phases come from dependencies, so nothing starts until what it relies on is verified. This is the calculator's real graph.

Calculator build graph: phase 0 calculation engine and app shell; phase 1 keypad and display; phase 2 history, web keyboard, offline and store builds; phase 3 launch readiness. PHASE 0PHASE 1 PHASE 2PHASE 3 N001 · foundationCalculation engine N002 · foundationApp shell & tokens N003 · featureKeypad & display N004 · featureHistory N005 · featureWeb keyboard & clipboard N006 · opsOffline & store builds N007 · launchLaunch readiness

Each node has a Definition of Done and its own implementation prompt in .raftly/prompts/.

What's inside

An operating system for building products with Claude

22 senior roles

From Head of Product Strategy to Principal SRE. Each decision is made by the discipline that should own it, and the reply says who decided.

Decision records & overrides

ADRs with confidence, reversibility and revisit triggers. Overrule Raftly and it records your call, the consequences, and moves on without re-arguing.

Evidence graph

Requirement → feature → screen → API → table → test → metric. Ask "why does this exist?" or "which requirements have no tests?".

Production gate

45 launch checks plus the commonly forgotten 20% (password reset, failed payments, restore drills…), each applied only when it fits your product.

Requirement change engine

Change a requirement and Raftly supersedes it with history kept, then shows the impact across features, decisions, data, tests and docs.

Learning firewall

Learns from your project without poisoning itself. "I hate rounded buttons" stays your preference, never a global rule.

Outputs

Everything lives in your repo

Plain Markdown and YAML you can review, diff and commit. No account, no backend, no lock-in.

  • Product, Design, Architecture, Build documents, plus Security, Data, GTM and Compliance-readiness when relevant
  • A visual HTML blueprint with the build graph, decision timeline, risks and readiness
  • Five product diagrams: architecture, user flow, screen map, data model and the critical-path sequence
  • A 12-slide deck with speaker notes, or your host's native deck tool
  • Self-contained prompts, one per capability, that any Claude session can run
my-product/
├── docs/
│   ├── PRODUCT.md  DESIGN.md  ARCHITECTURE.md
│   ├── DIAGRAMS.md       # 5 diagrams, Mermaid
│   ├── BUILD.md  SECURITY.md  DECISIONS.md
│   ├── decisions/ADR-001-client-only.md
│   ├── features/F001-calculation-engine.md
│   ├── prompts/001-calculation-engine.md
│   ├── blueprint.html
│   └── deck.md
└── .raftly/          # working state
    ├── state.yaml   requirements.yaml
    ├── build-graph.yaml  # dependency order
    ├── evidence.yaml     # intent → code → tests
    └── gate.yaml         # production gate
Install

Two commands. Then /raftly

Raftly installs into Claude Code once and works in every repository you open.

Inside Claude Code, add the marketplace and install the plugin:

# in Claude Code
/plugin marketplace add roszhan2684/raftly
/plugin install raftly@raftly

Then open any project and run:

/raftly I want to build ...

Requires Claude Code and Python 3.9+ (standard library only). If another tool already owns /raftly, use /raftly:raftly.

FAQ

Questions

Does Raftly send my code or plans anywhere?

No. Raftly has no backend, no account and no telemetry. Everything it writes stays in your repository. Research uses whatever web tools your Claude session already has.

Will it ask me a hundred questions?

No. It decides routine things itself and records important recommendations. It only asks about genuine business decisions such as pricing or target customer, batched into a single round.

Can I use it on an existing, half-built repo?

Yes. Raftly inspects what exists, records it as current, initialises its state without touching your code, and builds a recovery plan. It never starts with a rewrite.

What if I disagree with a recommendation?

Tell it. It records your decision as a founder override, keeps its original recommendation and the consequences, and proceeds consistently with your choice.

Does it certify my product as secure or compliant?

No. Raftly provides engineering readiness analysis and evidence-based guidance. It flags items that need professional review and never claims certification.

Is it free?

Yes. Raftly is open source under the Apache License 2.0. You need Claude Code to run it.

Have an idea? Type /raftly.

Get a researched, decided, buildable blueprint, and a partner that keeps it honest all the way to launch.