links: site | repo | project page
AI Blueprint moved from “repo under my GitHub account” into a more serious project surface this week.
The public repo now lives at
aiblueprinthq/ai-blueprint, the
site still lives at ai-blueprint.dev, and
create-ai-blueprint is up to 0.10.0 on npm. The old repo URL redirects, but
the new org matches the shape of the project better: this is becoming a workflow
product, not just a personal prompt pack.
The bigger product change is the CLI starting to exist for real. Setup still begins with:
npx create-ai-blueprint@latestBut the package now also exposes a shorter blueprint command. I kept that
global command intentionally narrow: it reports read-only project status. It can
find the project root, read Blueprint metadata, inspect current work, surface
open findings, look at Git state, warn about drift, and suggest the next action.
It does not install, update, commit, push, or run work on its own.
That boundary matters. A status command is useful precisely because it can be safe to run in the middle of a messy repo. It gives the next agent, or me after a context clear, a clean snapshot without becoming another automation lane too early.
The other important addition is GitHub Copilot support. AI Blueprint already had
Codex and Claude Code adapters, but Copilot belongs in the same workflow world
now that AGENTS.md is becoming a shared instruction surface. Version 0.10.0
adds Copilot support through the shared AGENTS.md and .agents/skills/
adapter files, and makes --all the default install mode. The old --both
flag still works as a deprecated alias.
This run also tightened the package as a real dependency: TypeScript output, Node 22+, package smoke checks, changelogged releases, and release tags that map back to the public repo.
The next step is the one I want to be careful with: turning this CLI foundation into deeper orchestration without letting it become “agent runs everything and hopes.” Status first. Automation later, with gates.