# Techtree 0.2.1

Improve one declared Skill under fixed conditions, and produce a comparison the
participant can check offline.

## Install

```sh
uv tool install --python 3.12 techtree==0.2.1
```

macOS or Linux · Python 3.12, provided by the installer · Docker required · Hermes 0.20.1+ for the plugin path

Release fingerprint: `sha256:482b3b97428cde5c7b9c35f9c9374b9cfc5ddb7b73ce9b6868b8536f4d014e2b`
Source revision: `a1b9c052c349a2acc40d0fe1ea206026efd0f8fc`

## First run

```sh
techtree doctor --climb hello-world-climb@1
techtree climb prepare hello-world-climb@1 --skill path/to/skill
```

Read the preparation output and run the exact one-time `techtree climb start`
command it prints. Nothing causing LLM token spend starts on its own.

## Data boundary

Techtree uploads nothing unless you publish a finished run yourself. Publishing uploads
the complete proof bundle — its manifests, signed report and receipts, cited documents,
and any optional execution record — while Episodes and Traces remain local. The network
returns a separate signed publication receipt acknowledging acceptance; it is not the
uploaded proof bundle.
The agent under test still makes model calls, and those go to the model provider
the Climb names, under that provider's policies. No Techtree account exists.

Documentation: https://techtree.sh/docs