Agent-assisted product initiatives · MIT · on npm
AI-assisted projects move fast — and rot fast. Blueprint is a working method for running a product initiative with an AI agent — research through deployment. The agent does the building; the method supplies the checkpoints and the paper trail. What ships is researched, prototyped, fact-checked, and documented.
$ npx @nino-chavez-labs/blueprint-cli init --name=my-initiative
plain language
Just the strip. Everything below goes deeper.
A method for doing product work with an AI carrying the heavy lifting. The AI researches, prototypes, and drafts the documents; Blueprint is the set of checkpoints that keeps the work verifiable.
AI-built work is fast and convincing — and that's the problem. Confident claims nobody verified, decisions nobody wrote down. Blueprint forces the receipts: every claim traced, every decision recorded, every stage checked before the next one starts.
Anyone shipping product work with an AI agent — solo or in a team. Solo, it's quality control: fast doesn't mean sloppy. In a team, it's the harder problem: everyone has their own way of keeping AI honest, and none of them line up. Blueprint is one way the whole team can share.
the thinking behind it — The kitchen I'm looking for ↗
how it works
Each stage produces artifacts that feed the next. The insight: prototype and documents are built simultaneously — the prototype tests the decision, the docs capture the rationale.
01 Research
competitive analysis, codebase, comparables
02 Design
define scope — what the prototype will and won't do
03 Prototype
HTML pages matching the real product
04 Fact-check
every claim validated against source
05 Document
strategy, feasibility, integration plans
06 Deploy
one portal — docs + prototype + chat
07 Iterate
stakeholder feedback, triaged
proof
Every page here was generated from a single blueprint.yml
by the same pipeline you’d run. The CLI is live on npm; the portal deploys from one config.
Blueprint runs on itself — the methodology, applied to productizing the methodology.
71gated updates
10decision records
15automated reviewers
13projects running Blueprint
7CLI commands
1config → whole portal
get started
Five steps. The choices you make are written out below — every command runs as printed.
Node 18+ and Claude Code. That’s it — there’s no server to stand up and nothing to host.
Generates a portal you own, validated against your config. --name is the only required flag — everything else defaults and is echoed back. Two starting points:
New idea — nothing exists yet. That’s all greenfield (the default) means.
$ npx @nino-chavez-labs/blueprint-cli init --name=my-initiative
Existing app — run it from that repo. That’s brownfield.
$ npx @nino-chavez-labs/blueprint-cli init --name=my-app --variant=brownfield --target=.
Set the three things that matter in blueprint.yml — the rest has sane defaults.
variant: brownfield # greenfield | midstream | brownfield
tier: 1 # deliverable sophistication 0–2
portal_pattern: A # A = platform portal · B = redesign reviewIn Claude Code, the stages are skills. Run them in order — each gate is enforced by a reviewer.
/blueprint-research → /blueprint-prototype → /blueprint-validate → /blueprint-docs → /blueprint-deploy
Push and your portal deploys. Verify conformance — doctor is honest about what it didn’t check.
$ npx @nino-chavez-labs/blueprint-cli doctor
Want the whole thing — stage by stage, every reviewer gate, on a real initiative? → Run your first initiative, or start at Learn.
reference
7 commands, all real. npx @nino-chavez-labs/blueprint-cli <command> — zero install.
Scaffold a new initiative and its portal — your variant, tier, and pattern choices validated before anything is written.
Run a stage gate: an automated reviewer that checks a stage is actually done. --list shows built-in + your own.
Per-stage effort/model configuration and telemetry — dial spend up or down.
One report across every project running Blueprint: which are current, which have fallen behind.
Move a project to the latest methodology version — preview first, dry-run by default.
Health check: loads the config, every reviewer, and the portal — and says what it didn’t check.
Set up the shared workspace a team of agents coordinates through. hive setup shows its plan first — nothing is provisioned until you pass --execute.
ecosystem
Updates flow down without breaking you; your fixes and gaps flow back up.
blueprint upgrade reads the methodology version your project is pinned to, shows what changed, and updates it — non-breaking, dry-run by default. blueprint fleet shows which of your initiatives have fallen behind.
$ npx @nino-chavez-labs/blueprint-cli upgrade
Hit a gap, a bug, or a feature need? File an amendment or RFC. Triage routes it automatically — substantial change → RFC first, bug-fix → straight PR.
further reading
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。