Your AI can build an app in a weekend.
Bansat makes sure it's one you can actually launch.

A blueprint-first framework and Agent Skill for Claude. Before any production code, your AI agent writes the plan a senior team would: product, architecture, security, privacy law, accessibility and every UI state. Then it stops and waits for your approval.

Install · See it work · 17 launch traps · What's inside · Cite

The bugs that sink AI-built apps aren't in the code

They're decisions nobody made. The demo works and the tests pass. Then:

Row Level Security is off, so anyone with your app's public key can read or change your tables.
Your fonts load from Google, handing every visitor's IP address to a third party before consent. A Munich court awarded damages for exactly this in 2022.
Hotjar or Clarity start recording before anyone has said yes.
The "we launched!" email goes to your waitlist with no unsubscribe link, no sender identity and no address.
The sign-up form never asks for age, so you may be holding children's data without parental consent. Under Nigeria's NDPA, a child is anyone under 18.
The privacy policy was written by an AI and describes an app that doesn't exist.

Bansat makes every one of those decisions explicit before the first commit, and checks them again before launch.

Bansat (بنسټ) is Pashto for "foundation": the part you lay before you build.

How it works

flowchart TB
    subgraph NEW ["Plan"]
        direction LR
        A["Your idea"] --> B["Intake<br/>20-question screen"] --> C["Risk tier<br/>Low · Medium · High"] --> D["docs/BLUEPRINT.md"] --> G1{"G1<br/>You approve"}
    end
    subgraph SHIP ["Build and ship"]
        direction LR
        P["9 build phases"] --> G2{"G2<br/>Data-ready"} --> G3{"G3<br/>Launch gate"} --> L["Launch"] --> G4{"G4<br/>30-day review"}
    end
    NEW --> SHIP

The gates are the point. At each one, Claude is instructed to stop and wait for you:

G1: no production code until you approve the blueprint.
G2: no real personal data, not even a waitlist email, until the privacy basics are live: the notice, processing records, consent capture, processor contracts, access-isolation tests and the ICO fee.
G3: no launch while any of the 17 blocking checks is red.

A skill can only instruct Claude, not force it. For a hard stop at G1, add the optional G1 hook: it refuses Claude's code edits until a person signs the blueprint.

Already built something? The retrofit track triages the live risks within 48 hours, reverse-engineers the blueprint from your code, and turns every gap into a fix with a deadline:

flowchart LR
    E["Existing or live app"] --> T["Day-0 triage<br/>11 checks in 48 h"] --> R["As-is blueprint<br/>from the code"] --> GR["Gap register<br/>with deadlines"] --> S["Fix sprints<br/>R1 → R4"]

Install in 60 seconds

Claude Code (terminal or VS Code). Install it once and use it in every project. Installs are pinned to a release, so the skill's instructions never change without you choosing to update:

# macOS / Linux
git clone --branch v2.1.0 --depth 1 https://github.com/bewithkhan/bansat ~/.claude/skills/bansat

# Windows (PowerShell)
git clone --branch v2.1.0 --depth 1 https://github.com/bewithkhan/bansat "$HOME\.claude\skills\bansat"

Start a new Claude Code session and type /bansat. To share it with your team, copy the folder into your repo at .claude/skills/bansat/ and commit it.

Updating: read the changelog, then switch the folder to the release you want (replace vX.Y.Z; on Windows, use "$HOME\.claude\skills\bansat" as the path):

git -C ~/.claude/skills/bansat fetch --depth 1 origin tag vX.Y.Z
git -C ~/.claude/skills/bansat checkout vX.Y.Z

Claude apps (web and desktop): add it as a skill in your Claude settings. See Use skills in Claude.

Any other agent: SKILL.md follows the open Agent Skills format, so skills-compatible agents can load it as it is. For Cursor, Copilot, Lovable, Bolt or anything else that reads a rules file, give the agent reference/bansat-framework.md and paste its Part 12.1 agent rules into AGENTS.md, .cursor/rules or your project knowledge.

Make G1 a hard stop (optional)

The skill tells Claude to stop at G1, but an agent can still skip a step. This Claude Code hook makes it a hard stop in a new project. Until a person signs the G1 row in docs/BLUEPRINT.md, Claude's file edits outside docs/, spikes/ and .claude/ are refused, and Claude can't sign the row itself. It needs Node.js 18 or later.

Copy the hook into the project:
```
mkdir -p .claude/hooks && cp ~/.claude/skills/bansat/hooks/bansat-g1-guard.mjs .claude/hooks/
```
On Windows (PowerShell): New-Item -ItemType Directory -Force .claude\hooks; Copy-Item "$HOME\.claude\skills\bansat\hooks\bansat-g1-guard.mjs" .claude\hooks\

Add this to the project's .claude/settings.json, merged with any hooks you already have:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Write|Edit|MultiEdit|NotebookEdit",
        "hooks": [
          {
            "type": "command",
            "command": "node",
            "args": ["${CLAUDE_PROJECT_DIR}/.claude/hooks/bansat-g1-guard.mjs"]
          }
        ]
      }
    ]
  }
}

When you approve the blueprint, fill in the G1 row yourself, for example | G1 Blueprint approved | Your Name | 2026-11-05 | |.

The hook guards Claude's file-editing tools, not shell commands, so treat it as a guard rail rather than a sandbox. Its tests run with node hooks/bansat-g1-guard.test.mjs.

See it work

In Claude Code, in an empty folder:

/bansat I want to build a revision app with an AI tutor for students in the UK and Nigeria

Bansat restates the idea and asks at most five questions, only where a wrong guess would be costly. It then writes docs/BLUEPRINT.md and a summary like this one (abridged):

Risk tier       HIGH: under-18s are a likely audience, plus AI features
DPIA            Required before any real data
Stack profile   D2 Progressive web app
Modules         Part 8 AI · 9.11 Marketing · 9.12 Children

Top 5 risks
 1  Children's data. The UK Children's Code is likely to apply, and Nigeria's
    NDPA treats anyone under 18 as a child: parental consent if you rely on
    consent, plus an age check.
 2  Student work sent to an LLM. Needs a DPA, no training on your data,
    redaction and a clear notice.
 3  The waitlist launch email. Needs consent evidence, one-click unsubscribe
    and sender details.
 4  Nigerian users. Past 200 in any six months you're a data controller of
    major importance and must register with the NDPC.
 5  AI cost abuse. Per-user quotas, a spend cap and a rate-limit screen (S13).

Assumptions     Web first · a waitlist before launch · no payments in the MVP
                13+ with a parental-consent flow                    (all flagged)
Open question   Will schools sign pupils up? Then you're their processor.
Missing for G1  Legal entity · ICO fee tier · AI provider

No production code until you approve this blueprint (G1).

Auditing an existing app works the same way: run /bansat audit this repo. The Day-0 triage comes first, marking each check pass, fail or unknown with its evidence: exposed secrets, missing RLS, public buckets, tracking before consent, unsafe marketing email, children's data, and whether anything was actually exposed.

The 17 launch traps

These are the failures that get vibe-coded apps fined, breached or rejected by the app stores. Bansat checks every one before you collect real data (G2), and again before launch (G3). The full table, with fixes and tests, is in §9.26 of the framework.

#	Trap	Why it bites
1	The sign-up form never asks for age	Children's data without parental consent or safeguards. Under Nigeria's NDPA, a child is anyone under 18.
2	Fonts from `fonts.googleapis.com` or any CDN	Visitor IP addresses go to a third party with no lawful basis. German courts have awarded damages.
3	Session replay or heatmaps load before consent	A consent breach, and recordings can capture personal data
4	A "we launched" email with no unsubscribe, sender identity or address	Breaches UK PECR, US CAN-SPAM and the NDPA, and fails Gmail and Yahoo sender rules
5	RLS off (or unscoped queries), or the secret key in the client	Anyone can read or write your database: a breach, with 72-hour deadlines
6	A template or AI-written privacy policy	It doesn't match what the app does, so it breaches transparency rules
7	Ad or analytics pixels in the root layout	Data shared with ad platforms before consent
8	User data sent to an LLM with no disclosure or DPA	Transparency, transfer and contract breaches, and possible training on your data
9	No way to delete an account or export data	A rights breach, and an App Store rejection
10	Personal data in logs or error trackers	Breaches minimisation and widens any breach
11	Pre-ticked boxes, or "by using this site you agree"	Invalid consent
12	A cookie banner with no "Reject all"	Invalid consent, and a top enforcement theme for regulators
13	Real customer data in preview or staging	A security and purpose-limitation breach
14	A public storage bucket for user uploads	Data exposure
15	ICO fee unpaid; NDPC status unchecked	ICO penalties of up to £4,350; missed duties for Nigeria's major-importance controllers
16	No breach plan	Missed 72-hour clocks
17	Dates of birth or ID scans kept "just in case"	A minimisation breach, and a honeypot for attackers

If you read one thing in this repo, read this table. Then check your own app against it.

Myth vs fact

Viral compliance threads get some of this wrong. Bansat sticks to what the law actually says:

The claim	What's actually true
"Your app can be fined ₦10m before its first sale."	₦10m or 2% of annual gross revenue, whichever is greater, is the most the NDPC may impose on a major-importance controller after an investigation. It isn't automatic. The risk is still real: targeting people in Nigeria counts as operating there, and the major-importance threshold is just 200 people in six months.
"Hotjar and Clarity record every keystroke, including passwords."	Not by default. The real failures are missing consent, relaxed masking and unmasked custom inputs.
"Anyone under 18 needs parental consent."	Under Nigeria's NDPA, yes, when you rely on consent. In the UK and EU, the age for consenting to online services is between 13 and 16. Build to your markets, or make the service 18+.

What's in the blueprint

One Markdown file: 15 parts and 4 appendices, about 2,800 lines. Claude reads only the parts each project needs.

Part	What it nails down
0 Intake & risk	20 screening questions, a risk tier, and which modules switch on
1 Product (PRD)	Problem, metrics, personas, MoSCoW features with IDs, Gherkin acceptance criteria
2 Non-functional	Targets with numbers: Core Web Vitals, uptime, recovery, security baseline, accessibility, cost
3 Architecture (TRD)	Pinned stack, environments, a context diagram, data flows, a processor register, secrets
4 Decisions (ADRs)	Baseline architecture decisions, accepted unless you override them
5 Data	Classification, a data dictionary, a Prisma 7 schema, RLS policies with isolation tests, retention
6 API	An endpoint inventory, RFC 9457 errors mapped to UI states, validation, webhooks, rate limits
7 Security	Auth, an RBAC matrix, 12 pre-filled STRIDE threats, headers and CSP, logging rules
8 AI features	A feature register with pinned models, data minimisation, evals, prompt-injection defences, EU AI Act labelling, spend caps
9 Privacy & compliance	27 sections, from lawful bases, consent and cookies to children, marketing, rights, breaches and DPIAs
10 Experience	Journeys, WCAG 2.2 AA, contrast-checked design tokens, and 17 mandatory UI states for every screen
11 Delivery	9 phases with exit criteria, a Definition of Done, RLS, privacy and E2E tests, CI/CD, incidents
12 Agent manual	Rules for `CLAUDE.md` or `AGENTS.md`, a feature protocol and an 8-step debugging protocol
13 Launch gate	37 go/no-go checks; any one of the 17 blocking checks stops the launch
14 Retrofit	Day-0 triage, reverse-engineering, a gap register, a review cadence and a portfolio register
A–D Appendices	Templates (DPIA, LIA, breach log and more), references, a glossary and 9 stack profiles

Why not a PRD template, or just ask the AI?

	PRD template	Asking the AI	Bansat
Product, architecture, security, privacy and UX in one plan	Product only	Patchy	✅
Gates Claude stops at until you approve	❌	❌	✅ G0–G4, plus an optional G1 hook
Laws cited to the section, and checked on a stated date	❌	Unverified	✅ 4 Nov 2026
Code samples tested before release	❌	Untested	✅
A track for apps that are already live	❌	❌	✅ Day-0 triage
Every UI state designed and tested on every screen	❌	Happy path only	✅ 17 states
The same standard on every project, team and tool	Copy and paste	Per chat	✅ One skill, one file

Tested, not just written

The code in the framework was run before release:

The Prisma 7 schema validates, with the connection in prisma.config.ts as Prisma 7 requires.
The Supabase RLS policies pass 14 tenant-isolation tests on a real Postgres database, and an audit query finds any table left without RLS.
The TypeScript UI-state pattern compiles in strict mode, and the build fails if a screen forgets a state.
The Playwright privacy test passes on a clean page and fails as soon as a page contacts a third party before consent.

These test files aren't in the repo yet. They're coming in an examples/ folder with CI, so you can check the claims yourself (issue #1). The G1 hook's tests are already here: node hooks/bansat-g1-guard.test.mjs.

Laws and standards covered (checked 4 November 2026)

UK GDPR and the Data Protection Act 2018, as amended by the Data (Use and Access) Act 2025
PECR (cookies and electronic marketing), with fines now at UK GDPR levels
The UK Children's Code (Age Appropriate Design Code)
EU GDPR and the ePrivacy rules, including the EU representative check
Nigeria Data Protection Act 2023 and the GAID 2025 directive
EU AI Act: transparency duties (Art. 50), AI literacy and the high-risk check
UK Online Safety Act 2023, for apps with user-to-user content
UK PSTI product security rules for connectable devices, and the European Accessibility Act
US: COPPA, CAN-SPAM, TCPA and state privacy laws
Pointers for India, South Africa, Kenya and Pakistan
Standards: WCAG 2.2 AA · OWASP ASVS · OWASP Top 10 for LLM Applications · STRIDE · RFC 9457 · RFC 9116 · RFC 2119

How this was checked: the legal sections were researched and checked against official sources (the legislation, regulator guidance and court decisions listed in Appendix B of the framework) with AI assistance, as of 4 November 2026. No lawyer has reviewed them yet. Laws change, so re-check anything you rely on before launch; see the disclaimer.

Works for any kind of project (9 stack profiles)

Profile	What it adds
D1 Web app (Next.js + Supabase)	The reference stack the examples use
D2 Progressive web app	Cache rules, offline and sync states, web push limits, and no background location on the web
D3 Mobile (Flutter, React Native, native)	No secrets in the binary, secure storage, store privacy labels, and Firebase rules as your RLS
D4 Wearables, BLE and connected devices	Device states, vendor clouds and transfers, consent from the people being monitored, PSTI
D5 Admin panels and CRMs	You're usually the client's processor, even when the client is abroad; exports are the biggest leak
D6 Research prototypes	Ethics approval, a data management plan, pseudonymisation at collection
D7 Websites and waitlists	A waitlist is already personal data, so the privacy basics come first
D8 AI agents and clinical prototypes	Synthetic data until reviewed, a medical-device check, a human in the loop
D9 Hardware with no personal data	Which parts apply, and when a sensor switches privacy on

Using Laravel, Rails, Django or a database without RLS, such as MySQL? The same rules apply: access is enforced in the data layer instead, and every controller re-checks authorisation. A dedicated profile for these stacks is on the way.

FAQ

Is this legal advice?

No. It summarises the law and regulator guidance as of 4 November 2026, cites the sections, and says when to get a qualified review: health, children, biometrics, worker monitoring or large-scale tracking. The legal sections were checked against official sources with AI assistance, and no lawyer has reviewed them yet.

Won't a blueprint slow me down?

It scales with risk. A low-risk project fills in only the essentials: intake, the core product spec, the stack, cookies and third-party requests, the launch traps, UI states and the launch gate. High-risk projects get everything, because that's where mistakes cost the most.

Do I need Claude?

No. The framework is one Markdown file you can fill in by hand or with any AI agent. Claude Code just does the most for you: it screens your idea, writes the blueprint and stops at each gate.

My team already has its own process.

Your process wins. If your repo's CLAUDE.md, AGENTS.md or docs set out their own planning stages or templates, Bansat follows them, maps its gates onto your stages, and fills the gaps it covers best: privacy law, security and UI states.

My country isn't covered.

The engineering half applies everywhere, and the laws table has pointers for other markets. Open an issue with your country's law and a link to an official source.

Can I use it commercially?

Yes. The text and templates are CC BY 4.0, so you can use and adapt them commercially as long as you keep the credit line. The code samples are MIT.

How do I keep a project's blueprint current?

It lives in your repo as docs/BLUEPRINT.md and changes in the same pull request as the code. Live projects get a monthly review and a quarterly launch-trap audit.

Cite

Use GitHub's Cite this repository button, or:

Khan, M. T. (2026). Bansat (بنسټ): A blueprint-first framework for building software (Version 2.1) [Computer software]. GitHub. https://github.com/bewithkhan/bansat

@misc{khan2026bansat,
  author       = {Khan, Muhammad Tufail},
  title        = {Bansat: A blueprint-first framework for building software},
  year         = {2026},
  note         = {Version 2.1},
  howpublished = {\url{https://github.com/bewithkhan/bansat}}
}

Contributing

Bansat gets better with every project it meets. Contributions are welcome:

A law changed? Open an issue with a link to the official source.
A stack that isn't covered? Propose a profile for Appendix D.
A launch trap that bit you? Describe what happened and how you fixed it.

Keep the credit line in anything you copy or adapt (CC BY 4.0).

Author

Muhammad Tufail Khan: PhD researcher in AI and IoT for healthcare at the University of the West of Scotland, based in Glasgow.

Licence

Text and templates: CC BY 4.0, full text in LICENSE. You can use, adapt and share them, including commercially, if you keep this credit: "Bansat (بنسټ) by Muhammad Tufail Khan, https://github.com/bewithkhan/bansat, CC BY 4.0".
Code samples: MIT (see LICENSE-CODE).
NOTICE sets out which licence covers what, with the credit line.

Disclaimer

Bansat summarises laws and regulator guidance as of 4 November 2026. It is not legal advice. The legal sections were checked against official sources with AI assistance and have not been reviewed by a lawyer. Check the current rules before you launch, and get a qualified review for high-risk processing.

If Bansat saves you from one bad launch, star the repo so the next builder finds it before theirs.
Built by Muhammad Tufail Khan · LinkedIn · GitHub

Your AI can build an app in a weekend.Bansat makes sure it's one you can actually launch.