FOUNDRY

Foundry is a Claude Code plugin that turns a one-line brief into a designed, architected, implemented and tested web application, asking the founder at most 7 questions plus 3 follow-ups. It runs inside your logged-in Claude Code session and writes everything to .foundry/ in your app repo.

Install

Base tools (Git, Node LTS, Python 3.11+), then Claude Code:

winget install --id Git.Git -e; winget install --id OpenJS.NodeJS.LTS -e; winget install --id Python.Python.3.12 -e   # Windows
brew install git node python@3.12                                                                                    # macOS
npm install -g @anthropic-ai/claude-code

Then the plugin, from GitHub:

claude plugin marketplace add ryktara/Foundry
claude plugin install foundry@foundry

Dev path (a local checkout; the marketplace is named foundry from a checkout too; it was foundry-local before 1.0):

claude plugin marketplace add /path/to/foundry
claude plugin install foundry@foundry

In a claude session, run /foundry-setup; done when the last line starts with setup: ready. Step by step, with troubleshooting: docs/INSTALL-WINDOWS.md, docs/INSTALL-MAC-LINUX.md.

Quickstart

Full guide: docs/QUICKSTART.md (Arabic). From an empty app folder:

/foundry "<one line: what the business is, where, what you already know>"
/foundry-build
/foundry-resume
/release

/foundry plans (phases 0 to 10), /foundry-build implements the next 4 tickets (--n 10 for more), /foundry-resume continues in a new session, /release produces deploy files and a smoke-tested build.

Packs available

Pack	Status	Covers
restaurant-pos	complete	restaurants and cafes: tables, kitchen display, offline tills, bilingual receipts
retail-pos	complete	shops and supermarkets: barcodes, stock, returns, e-invoicing
trading-app	complete, regulated	brokerage and trading apps: orders, portfolio, KYC; release needs a licence confirm
generic	fallback	any brief no domain pack matches with confidence 0.7 or more

Write a new pack in a day: docs/PACK-AUTHORING.md.

Pipeline

Phases 0 to 15; gates, inputs and outputs per phase: docs/pipeline.md.

+-----------+   +-----------+   +---------------------+   +-----------+   +-----------+
| 0 brief   |-->| 1 match   |-->| 2 grill             |-->| 3 PRD     |-->| 4 domain  |
|           |   |   pack    |   |   <=7 + 3 questions |   |           |   |           |
+-----------+   +-----------+   +---------------------+   +-----------+   +-----------+
                                                                                |
+-----------+   +-----------+   +---------------------+   +-----------+         |
| 8 screens |<--| 7 design  |<--| 6 data + API        |<--| 5 archi-  |<--------+
|           |   |   system  |   |                     |   |   tecture |
+-----------+   +-----------+   +---------------------+   +-----------+
      |
      v
+-----------+   +-----------+   +------------------------------------------------+
| 9 threats |-->| 10 tickets|-->| 11-12 build loop, per ticket                   |
| + comply  |   |   (DAG)   |   |   builder -> finisher -> 3 reviewers           |
+-----------+   +-----------+   |   (code, ui, security)                         |
                                +------------------------------------------------+
                                                   |
                +-----------+   +-----------+   +-----------+
                | 15 handoff|<--| 14 release|<--| 13 wizards|
                +-----------+   +-----------+   +-----------+

Cost expectations

Measured on the reference app (restaurant-pos, 49 tickets, P8–P11 transcripts), subscription usage priced at public list prices: sonnet for builder, finisher and reviewers, haiku for ui-review, the parent's own turns at the opus tier. Full table: docs/examples/restaurant-pos/COST.md.

Measure	Value
Per feature ticket, median	$21.20
Per compliance ticket, median	$10.16
Per integration ticket, median	$5.21
Full reference app, total (49 tickets)	$1,059.64 (sonnet $473.57, haiku $7.67, parent $578.39)
Wall time, sum of ticket windows	46 h
Design pipeline, phases 0–10 (retail-pos run)	$4.29, under 30 min

Your own numbers, from the app folder: python <foundry>/scripts/foundry.py metrics report --by-model.

Limits

• Web-first: every app is nextjs-pwa + hono + postgres. Expo mobile is on the roadmap.
• Subscription-only: no API key, no non-interactive CLI, no headless mode. Model evals run through /foundry-eval in a session.
• Human-only steps (provider keys, domains, licences) are handed to you as wizards; the feature ships stubbed until you run it.
• Regulated packs block release until a named person runs release confirm --by <name>.
• DoD runs (Playwright, embedded Postgres, parallel test runners) need a quiet machine; close other heavy work while building.

Roadmap

• Expo stack for native mobile apps.
• More domain packs.
• Preview environments per ticket.
• Cross-project intelligence: lessons from one app's build feeding the next.

Layout

.claude-plugin/   plugin manifest and marketplace (foundry)
skills/           one SKILL.md folder per skill and subagent
packs/            domain packs, packs/index.csv, packs/README.md (schema 2.0)
data/             CSV knowledge the scripts query
schemas/          JSON Schema draft 2020-12
scripts/          foundry.py CLI, wrappers, tests
hooks/            session start, guards, metrics, stop check
evals/            briefs, expected outcomes, fixtures, thresholds, run.py
docs/             install, quickstart, pipeline, models, pack authoring, orchestrator, examples
CHANGELOG.md      step-by-step history (P0 to P11)

Licence

Copyright (c) 2026 Al Sadq IT Solutions LLC. All rights reserved. See LICENSE.

Source-available, not open-source: you may read and evaluate the code here; any other use needs the owner's written permission until Al Sadq IT Solutions LLC says otherwise. See CONTRIBUTING.md and SECURITY.md.