Product · HumanMaxxing CoreDraft
The framework, and the skills that run it.
Open source. Build a team of AI specialists from your own repo, run every ticket through the same gated workflow, with a trusted advisor between you and the team.
Build your team
One skill surveys your codebase and proposes the specialist agents it actually needs: not a generic roster, but reviewers and builders with your real commands, paths and domains filled in. You approve the roster before a single file is written.
How the team gets built
Machine · runs without you
Survey
The skill reads your repo the way a new senior hire would. Read-only.
- Frameworks, languages, test runners and the commands that run them
- Auth boundaries, API surface, migrations, infrastructure as code
- The business domains with enough code to deserve their own expert
machine runs human reviews
Archetypes it can propose: Code reviewer · Test engineer · Security reviewer · Performance reviewer · API engineer · Frontend builder · Migration engineer · Docs writer · Domain expert (one per domain) · Infra expert (if IaC is detected)
Run your team
One command takes a ticket to a planning session. It creates the worktree, opens the agent on the right model, and launches the kickoff workflow, which puts your team to work and stops at your two review points.
maxx-start 12312One ticket, from kickoff to plan.maxx-start 12312 4435 5345Work on multiple related tickets together, optimizing time and tokens.One change, start to finish
Machine · runs without you
Start
The launcher creates a worktree for the ticket, opens the agent in it, and runs the kickoff skill. It reads the ticket and checks for related open issues worth bringing in.
- Worktree and branch named from the ticket
- Already-shipped and in-flight guard before any work starts
- Related issues proposed for the same session, once
machine runs human reviews
Your trusted advisor
Between you and the team sits one agent whose job is to protect your attention. It classifies the work, asks its questions before dispatching anything, watches the specialists for rule violations, and verifies their results independently before bringing them to you.
Asks before it acts
Clarifying questions come first, in one batch, before anything is dispatched.
Dispatches the right specialist
Each piece of work goes to the agent built for it, on the cheapest model that can do it.
Verifies independently
A specialist’s “done” is re-checked before it reaches your gate.
Never decides for you
It cannot file tickets, add trust labels or skip a testing gate; it brings you one question at a time.
Keeps the record
Your decisions are dated and kept with the plan.
Learns from your sessions
Its instructions are mined from how you actually work, corrections included.
Questions
Which agents does it work with?
Built and run daily on Claude Code. The skills, templates and review structure are plain Markdown and shell, so other agents can follow them; hook enforcement is Claude Code–specific today.
What does “open source” cover?
Everything on this page: the team-building skill, the advisor, the kickoff, review and ship skills, and the templates. The hosted products on the other Product pages build on top of it.
How long does setup take?
One paste into your agent. It clones the framework, interviews you about how you work, proposes your team, and stops for approval before writing anything.