start-from-here

Open Agent Skill · MIT · zero dependencies

Plan first. Then build.

start-from-here interviews you and writes the approved project documents your AI coding agent needs — PRD, architecture, database, security, tests, legal and AGENTS.md — before a single line of code is written.

Install in your project folder

npx skills add ilramdhan/start-from-here

No Node 22.20+? Use our zero-dependency installer (Node 18+):

npx start-from-here@latest install

Then tell your agent:

Use the start-from-here skill to plan my new app.
All install options View on GitHub
  • 30+ AI tools supported
  • 30+ document templates
  • 3 tiers: Lite, Standard, Enterprise
  • 0 lines of app code written by the skill

Why plan before you prompt?

“Vibe coding” — asking an AI to build an app straight from a one-line idea — feels fast. But the agent has to guess.

The problem

  • Silent guessesScope, users, data model, permissions, security rules and conventions are all invented on the fly. Every guess becomes code.
  • Expensive reworkWrong guesses surface late — after the schema, the auth flow and the UI are built on top of them.
  • No shared memoryA new session, a new model or a new teammate starts from zero, because nothing was written down.

The solution

  • It interviews youShort rounds of numbered questions, each with a labeled recommended default.
  • It recommends, you decideNo stack yet? You get 2–3 options with trade-offs and one marked Recommended.
  • It writes consistent docsShared IDs link every document, and a final audit cross-checks them.
  • It never makes things upUnknowns become [TBD], ideas become PROPOSAL, guesses become ASSUMPTION.
  • It never writes app codeIts job ends with approved documents, a readiness checklist and a first implementation prompt.

How it works

Eight phases with explicit gates. The skill stops at every gate and waits for your answer.

Diagram of the eight phases. Arrows show the order; gates marked STOP wait for your approval. STOP STOP STOP STOP 0123 4567 Orient Calibrate Discover Recommend Confirm brief Generate Audit Handoff
  1. 0
    Orient

    New project, existing repo, resume or doc update? Can it write files here?

  2. 1
    Calibrate STOP · you approve

    Language, experience, idea and size. Proposes a tier and the doc list.

  3. 2
    Discover

    Rounds of up to 7 questions, each with a recommended default.

  4. 3
    Recommend STOP · you approve

    2–3 stack, hosting or license options with trade-offs.

  5. 4
    Confirm brief STOP · you approve

    Writes docs/DISCOVERY.md and shows a short summary.

  6. 5
    Generate STOP · you approve

    Writes docs in dependency order, in review batches.

  7. 6
    Audit

    Cross-checks scope, names, stack, roles, NFRs, legal and links.

  8. 7
    Handoff

    Readiness checklist, open questions and a first implementation prompt.

  9. →
    You build with any agent

Resumable

Progress is saved to docs/.start-from-here/STATE.md after every step.

Brownfield-aware

In an existing repo it inspects manifests, CI and docs first.

Works without a filesystem

In chat-only tools each document arrives as a Markdown block.

Any language

Chat in one language and write the docs in another.

What you get

These files are generated inside your project. Which ones you get depends on the tier and your answers.

Tier badges: L Lite S Standard E Enterprise ? conditional

Product

docs/DISCOVERY.mdLSE
Interview record: answers, assumptions and decisions.
docs/README.mdLSE
Docs index with a Draft / Approved status table.
docs/product/PRD.mdLSE
Vision, personas, features with acceptance criteria, success metrics.
docs/product/APP_FLOW.mdL?SE
Screens, commands or endpoints, user flows and states.
docs/product/DESIGN_SYSTEM.mdS?E?
Design tokens, components, states and accessibility.

Technical

docs/technical/TRD.mdSE
Feature-to-technical mapping and measurable NFRs.
docs/technical/ARCHITECTURE.mdLSE
C4 context and container diagrams, stack table, data flow.
docs/technical/DATABASE.mdL?S?E?
Entities, ERD, schema, access rules, migrations and retention.
docs/technical/API.mdS?E?
Contract conventions, endpoints, RPCs or CLI, errors and versioning.
docs/technical/SECURITY_ARCHITECTURE.mdLSE
STRIDE threat model deepened by a design-time threat catalog, authN/Z matrix, secrets and privacy.
docs/technical/TESTING.mdLSE
Test strategy and an acceptance matrix.
docs/technical/OPERATIONS.mdSE
Environments, CI/CD, deploy and rollback, SLOs, backup and incidents.
docs/technical/IMPLEMENTATION_PLAN.mdLSE
Milestones and ordered tasks, each with a definition of done.
docs/decisions/SE
Architecture decision records (MADR).

Engineering & AI governance

AGENTS.mdLSE
Universal instructions for any AI coding agent; links every doc.
CLAUDE.md · GEMINI.md · copilot-instructions.md?
Optional thin pointer files to AGENTS.md.
docs/engineering/CODE_STYLE.mdLSE
Conventions, project structure and verified commands.
docs/engineering/AI_GUIDELINES.mdLSE
AI coding dos and don'ts.
docs/engineering/GIT_WORKFLOW.mdSE
Branching, Conventional Commits, PR review and releases.

Drafts for review by qualified counsel — not legal advice.

LICENSELSE
Full official license text — never paraphrased.
docs/legal/TERMS_OF_SERVICE.mdS?E?
Terms of service, if you have accounts or a SaaS.
docs/legal/PRIVACY_POLICY.mdL?S?E?
Privacy policy, if you process any personal data.
docs/legal/COOKIE_POLICY.md?
Cookie policy, if a website uses tracking.
docs/legal/COMPLIANCE.mdS?E
Regulatory register: regimes, controls and owners.
docs/legal/MASTER_SERVICES_AGREEMENT.mdS?E?
Master services agreement skeleton, if you sell to businesses.
docs/legal/DATA_PROCESSING_AGREEMENT.mdS?E?
Data processing agreement skeleton (GDPR Art. 28), if you process customers' personal data.
docs/legal/BUSINESS_ASSOCIATE_AGREEMENT.mdS?E?
HIPAA business associate agreement skeleton, if you handle health data for covered entities.

Open source & GitHub

README, CHANGELOG, CONTRIBUTING, CODE_OF_CONDUCT, SECURITY, SUPPORT, ACCESSIBILITY, CODEOWNERS, PR and issue templates.

The threat model lives in docs/technical/SECURITY_ARCHITECTURE.md. Root SECURITY.md is reserved for GitHub's vulnerability-disclosure policy.

Tiers

The skill proposes a tier from what it learns, and you confirm it.

L Lite

For
Solo, indie, prototype, hackathon, internal tool
Interview
≈ 3 rounds, 15–25 questions
Review
3 grouped batches

S Standard

For
Startup MVP, open-source project, small-team product
Interview
≈ 5 rounds, 35–60 questions
Review
6 batches

E Enterprise

For
Regulated data, multiple teams, SLAs, multi-tenant, audits
Interview
≈ 7+ rounds, 60+ questions
Review
One document per review

Any regulated-data signal makes the skill propose Enterprise unless you object.

Experience modes

Chosen in the first round, or inferred from how you write.

Beginner

Explains every technical term in one plain sentence.

Intermediate

Short explanations of less common terms.

Expert

Terse. Shows trade-offs and alternatives.

Not sure what something means? Say “explain ERD” (or any term) at any time.

Install on your platform

Pick your tool. Every command comes from the official platform guides in this repository.

Vercel skills CLI — installs for every agent it detects

The fastest route when you have Node 22.20 or newer.

Needs: Node ≥ 22.20

npx skills add ilramdhan/start-from-here
npx skills add ilramdhan/start-from-here -g            # global
npx skills add ilramdhan/start-from-here -a cursor     # one agent

Full guide

Compatibility

Native skill support means the tool loads SKILL.md folders on demand.

PlatformSupportInvocation
Claude CodeNative skill + plugin/start-from-here
Claude.ai / DesktopNative skill (upload)Natural language
Claude API / Agent SDKNative skillcontainer.skills
OpenAI CodexNative skill + plugin$start-from-here
ChatGPTNative skill (upload)@start-from-here
OpenAI API / Agents SDKNative skillSkills(from_=…)
Gemini CLINative skill + plugin/start-from-here
Google AntigravityNative skillNatural language
Gemini app (Gems)Prompt-only“start”
GitHub CopilotNative skill/start-from-here
CursorNative skill/start-from-here
Windsurf / Devin DesktopNative skill@start-from-here
Cline · Roo Code · Kilo CodeNative skillNatural language
JetBrains JunieNative skill/start-from-here
KiroNative skillNatural language
OpenCode · Goose · Amp · Augment · Qwen Code · Mistral Vibe · ZedNative skill/start-from-here
Continue · Trae · OpenHands · Factory DroidPartial / unverifiedNatural language
Aider · Amazon Q · JetBrains AI Assistant · JulesRules-file fallback“start”
Ollama · LM Studio · Open WebUI · Le Chat · DeepSeek · Grok · PerplexityPrompt-only“start”

Agents whose paths are known only from third-party sources are marked UNVERIFIED. Full matrix and decision tree

Usage & quick commands

Say these at any time.

Start it

ToolSay
Claude Code, Copilot, Cursor, Junie, Zed, Gemini CLI/start-from-here
Codex, Open WebUI$start-from-here
Windsurf, ChatGPT Skills@start-from-here
Anything elseUse the start-from-here skill to plan my new app.
Pasted prompt, Gem, Custom GPTstart

Add your idea on the same line to save a round.

Check your docs any time

Finds missing files, leftover placeholders and broken links in a generated project.

npx start-from-here@latest check .
CommandEffect
startBegin at phase 0.
resumeContinue where you left off.
statusShow phase, tier, document status and open questions.
use your recommendationsAccept every recommended default in this round.
skip to generationFill gaps with labeled assumptions.
audit docsRun only the consistency audit.
update <doc>Edit that doc and every dependent doc.
add <doc>Generate one extra document.
explain <term>Plain-language explanation with one example.
change tier <…>Recompute the document set.
handoffJump to the handoff phase.
nextChat-only mode: send the next batch.

Example session

A beginner plans a booking app in Claude Code. Abridged.

You

/start-from-here a booking app for my yoga studio

Agent

Phase 0 — new project…

You

beginner, public, at the studio for now, just me.

Agent

Proposed tier: Lite…

…

…

Agent

Phase 7 — Handoff…

The first implementation prompt

The skill stops at handoff. Paste this into any coding agent.

Read AGENTS.md and every document it links. Summarize the approved MVP, architecture,
data and permission rules, coding conventions, and acceptance criteria. List blocking
contradictions or [TBD] items. Then propose a plan for Milestone 1, Task 1 from
docs/technical/IMPLEMENTATION_PLAN.md, with files to create and how each acceptance
criterion will be verified. Do not write code until I approve the plan.

Frequently asked questions

Does it write any code?

No. It writes planning documents only. Small illustrative snippets inside docs (a config key name, a command already verified in your repo) are allowed, but no application code, packages, migrations or deployments.

Which AI model do I need?

Any capable model. The skill is model-agnostic plain Markdown that follows the open Agent Skills standard, and it also ships as a single pasteable prompt for chat-only tools and local models.

Can I use it on an existing project?

Yes.

Will it invent facts or claim compliance?

No.

Do I have to answer every question?

No.

What if my session ends halfway?

Progress is saved to docs/.start-from-here/STATE.md after every step, and a short project digest (DIGEST.md) keeps names, IDs and decisions. Open any tool with any model and say “resume”. In chat-only tools, save the state and digest blocks yourself and attach them.

Is my data sent anywhere?

The skill makes no network calls of its own.

Which install method should I pick?

Terminal agent with Node 22.20+: npx skills add. Older Node: npx start-from-here install. Claude Code, Codex or Gemini CLI with auto-updates: the plugin or extension. Chat apps: upload the zip, or paste the prompt file.

How do I update it?

Re-run your install command.

Is it free?

Yes. It is open source under the MIT license. You only pay for the AI tool you already use.

My AI has a small context window. Will it work?

Yes — use session mode. Say “session mode” and the work is split into short work packages, one per fresh session, connected by a project digest (DIGEST.md) and a handover file (HANDOVER.md) in docs/.start-from-here/ that keep names, decisions and IDs consistent. The skill recommends it by default for Enterprise runs, more than ~20 documents, small or local models and chat-only tools, and checkpoints at every review gate so later documents stay as good as the first.

What is start-from-here?

An open, MIT-licensed Agent Skill that interviews you with recommended defaults, recommends a stack, gets your brief approval, and then generates a complete set of pre-development documents before any code is written.

What documents does it generate?

PRD, app flow, design system, TRD, architecture with ADRs, database and ERD, API, security threat model, testing, operations, implementation plan, code style, git workflow, AGENTS.md, legal drafts (terms of service, privacy, cookies, compliance, and conditional MSA, DPA and BAA) and open-source files.