From the lab · a fully-owned skill

Your AI Usage Card, in one paste.

A prompt you run in your own Claude Code. It audits your real setup and habits — verification, autonomy, parallelism, memory, cache — and prints a calibrated card: your level (L0–L5), your ≈ adoption step, and your one next unlock. Then the ladder below turns that level into a path.

~2 min 100% local nothing sent unless you choose no secrets · no paths · no names

Open Claude Code in any project, paste the whole prompt, let it run (~1–2 min). It works entirely on your machine and sends nothing automatically. At the end it prints your card plus a USAGE_CARD_JSON= line — sharing that line back via the form on this page is optional.

claude — any project — paste & run

    
the whole product is this prompt — that's the point

What comes back

A real card — this one is mine, measured, not self-declared.

LevelL4 — Harness Engineer
≈ ChernyStep 3 — Supervised Autonomy · next unlock: cost-lanes + exception-monitoring
WhyAuthored, git-versioned hook lifecycle with an exit-2 stop verification gate; agent teams in use; scheduling evidenced.
Operatesverify: enforced · autonomy: wide+guardrails · steering: low · reviews: summaries
Parallelroutine-parallel — 549 subagent + 570 background runs, worktrees, 4 named teams

measured 2026-07 · sharing the one-line JSON under the card is optional — one anonymous row, distributions only

Got your level? Find your rung.

Every rung expands into a hand-picked path — what to read, what to build next, and where I can help. That's the accompagnement: your card names the rung, the ladder hands you the path.

L5 Force Multiplier

Multi-agent at scale, evals & quality gates, tooling the whole team runs on.

Build

Evals over your own harness; share it as a team standard.

Work with me

CTO & longer missions — spread it across the org.

L4 Harness Engineer

You built and operate your own system: hooks, enforced verification, orchestration, living memory.

Build

Evals + exception-only monitoring; cost lanes per model.

Work with me

Half-Day Deep Dive — push toward team-scale.

L3 Agent Builder

Delegates real work to subagents & workflows, habitual verification, some machinery of your own.

Build

Your first lifecycle hook; give one agent living memory.

Work with me

1:1 Session or 4-Pack — author your harness.

← most engaged daily users centre here →
L2 Power User

Tuned settings, skills & MCP in real use, some verification — still mostly hand-driving.

Build

A self-verification loop: make Claude prove its work before you read it.

Work with me

Power Hour — one concrete win on your setup.

L1 Daily Helper

Regular interactive use, little authored config, you eyeball the output.

Build

Plan mode before every non-trivial task; a real CLAUDE.md.

Work with me

Power Hour — set the foundations right.

L0 Explorer

Occasional chat-style use, default setup.

Build

Let it edit a real file today — agentic, not chat.

Work with me

Foundations session — the mental model first.

The story behind it: why measuring behavior beats counting installed tools, how the ladder was calibrated on a room of 50 engineers, and what a "next unlock" really is — the full working note lands in the Lab shortly.

Share your card — optional

If you ran the prompt, I'd genuinely love to see where you landed. Paste your card — including the USAGE_CARD_JSON= line — and submit. One anonymous row, no email, no account.

Privacy, plainly: the prompt runs 100% locally and sends nothing automatically. The card carries no secrets, no paths, no employer names — and sharing it back is a choice, never automatic. Shared cards only ever appear as distributions, never individually.
No tracking. No account. You choose what to share.