proxify-dojo 2026.7.22.1

The Proxify Dojo — a guided, compounding campaign that turns a developer from 'uses Claude as a chatbot' into an agentic engineer. One command to start.

🥋 Proxify Dojo

Turn "uses Claude as a chatbot" into an agentic engineer.

A guided, compounding campaign — one challenge at a time, each building on the last. The CLAUDE.md you write in challenge 1 is still load-bearing at the finale.

uvx proxify-dojo start 1.6

That one line accepts your challenge, sets up a coached Claude session, and drops you in.

Or just run dojo to browse — an honest navigator of the whole campaign:

---

What it is

The dojo is one campaign, not a bag of exercises. Each challenge is session-scoped (≈10 minutes), ships one real artifact you keep, and teaches one named agentic-engineering capability. You climb a ladder, you don't grind isolated katas.

proxify-dojo is the front door: a single, self-contained CLI that runs the whole loop.

uvx proxify-dojo start 1.6
   │
   1. creates agentic-1.6-<you> from the template — under YOUR account (you always have access)
   2. clones it locally
   3. projects the coaching rig (the coach, the sandbox, the proof) into the folder
   4. launches Claude as a one-step-at-a-time coach
   │
   └─ you write the files → dojo check → dojo submit → dojo next

The loop

step	command	what happens
build	dojo start 1.6	a coach walks you through writing the files — it won't write them for you
prove	dojo check	two fresh Claude instances prove your work — the only honest test
keep	dojo submit	your work is pushed and tagged complete
next	dojo next	your own files carry forward into the next challenge

dojo (bare) shows the campaign map. dojo progress shows your character sheet. dojo doctor tells you exactly what's wrong if something's off — with the fix.

Install

# zero-install, one-shot:
uvx proxify-dojo start 1.6

# or install it:
uv tool install proxify-dojo
dojo start 1.6

Requires Claude Code and the GitHub CLI on your PATH. Run dojo doctor to check.

How it's built (the design that makes it scale)

One artifact, one version. The wheel ships everything a learner runs — the rig, the proof harnesses, the sandbox, the GitHub driver, the VS Code extension, AND the baked content snapshot. There is no separate content channel: new or edited content reaches learners in a wheel release (CalVer). In a source checkout the same content is projected live from stations/ on every launch, so authoring needs no rebuild step — the wheel just bakes a snapshot of that identical projection.

• Rig-in-package. The coach prompt, the proof harness, and the directory sandbox are owned by this package and projected into your challenge folder at runtime. Fix the coach once, everyone gets it next start — challenge repos never carry a stale rig.
• GitHub-native onboarding. dojo does accept → clone → submit itself, creating the repo under your own account — no org-invite dance, no extra extensions to compile.
• Latest by default. uvx proxify-dojo re-resolves the newest release each run; a persistent uv tool install is nudged by dojo start when a newer wheel is out.

Commands

dojo                  browse the campaign in an interactive navigator
dojo start <slug>     accept + set up a challenge and open the coach
dojo resume [<slug>]  re-open the coach
dojo check [<slug>]   prove it with fresh Claude instances
dojo restart [<slug>] wipe your work back to a clean slate
dojo submit [<slug>]  push + tag your completion
dojo next             carry your work into the next challenge
dojo list             the campaign map
dojo show <slug>      one challenge in detail
dojo progress         your journey + capabilities earned
dojo doctor           preflight, with exact fixes

Operator verbs (author, publish, cohort, …) live in the same binary and unlock with dojo login for dojo admins.

Develop

uv venv && uv pip install -e '.[dev]'
uv run pytest
uv run dojo list

Try the whole loop with no GitHub using local mode:

uv run dojo start 1.6 --no-launch    # default is a local copy — no GitHub needed
uv run dojo list

---

_{Built by the Proxify Agentic Engineering team · MIT}