Getting Started¶

Four commands¶

git clone --recurse-submodules https://github.com/fwornle/coding ~/Agentic/coding
cd ~/Agentic/coding && ./install.sh
source ~/.zshrc          # or ~/.bashrc
cd ~/my-project && coding

Two questions¶

The installer asks whether bare agents should be observed too (Enter = no, only sessions started via coding), and which tier to install:

harness needs no Docker. learning and up add session logging, online learning and the knowledge viewer, and need Docker running. Change it any time with coding-features profile <tier> or in the dashboard. Details: Installation.

The first launch in a repo¶

With a learning tier, the first coding in a repo asks once where its learned data goes — Enter creates a private <repo>-history, a teammate's URL shares theirs, skip keeps it local. See Per-Repo Tenancy.

Check it worked¶

The status line under the agent shows [🏥●] green when the system is healthy. For the full picture open the dashboard at localhost:3032.

Then: Configuration for LLM provider keys, or Verify & Repair if something is not green.

What you need¶

macOS, Linux, and Windows through WSL (with systemd enabled) are supported.

Installing¶

git clone --recurse-submodules https://github.com/fwornle/coding ~/Agentic/coding
cd ~/Agentic/coding && ./install.sh
source ~/.zshrc

The installer first lists everything it would touch (./install.sh --dry-run stops there), then asks two questions — agent scope and tier — and installs only what that tier needs:

Unattended: ./install.sh --yes --features=learning. The full walkthrough with every prompt is on Installation.

Your first session¶

cd ~/my-project
coding                  # the best available agent
coding --claude         # or pick one: --copilot, --opencode, --pi
coding --project DIR    # start in another directory

The agent opens inside a tmux session with the shared status bar along the bottom:

Launching the agent's own CLI directly skips session logging, knowledge injection and health monitoring — prefer coding.

Where what it learns goes¶

Everything learned in a repo is kept in that repo's .coding/ folder — session logs, observations, insights, knowledge-graph slice, token usage. On the first launch you decide whether that folder is a private <repo>-history repo (shareable with teammates through git), or local only. Commits happen automatically; pushing is always your call:

coding sync              # what each learning repo has to push / pull
coding sync --push       # asks, then pushes

Group repos into teams in Dashboard → Teams; the knowledge viewer follows that selection. See Per-Repo Tenancy and Teams & Shared Learning.

Checking health¶

The status line's [🏥●] is the summary. The dashboard at localhost:3032 shows each database, service and process, and Run Verification re-checks everything on demand:

From a shell:

coding-features status                 # which features are on, and why any are off
curl -s localhost:3034/health          # the health coordinator

Verify & Repair works through each service when something is red.

How the pieces fit¶

coding is a launcher. It resolves which features are on (~/.coding/features.yaml, written by the installer), starts or reuses what those features need, prepares the repo's learning checkout, and opens the agent CLI inside a tmux session carrying the shared status bar.

With only harness features on, the launcher never starts Docker.

Prerequisites¶

brew install git node jq tmux gh
brew install --cask docker        # learning tiers only

sudo apt update && sudo apt install -y git nodejs npm jq tmux gh
curl -fsSL https://get.docker.com | sh   # learning tiers only

Install¶

git clone --recurse-submodules https://github.com/fwornle/coding ~/Agentic/coding
cd ~/Agentic/coding
./install.sh            # asks: agent scope, tier
source ~/.zshrc         # or ~/.bashrc

Everything is reversible with ./uninstall.sh. The full flow, with every prompt, is on Installation.

What each tier installs¶

Switch later with coding-features profile <tier>, one feature with coding-features set <feature> on|off, or in Dashboard → Features.

Verification¶

• Status line — [🏥●] green = healthy; each other badge belongs to one feature and is absent when that feature is off.
• Dashboard — localhost:3032, Run Verification for an on-demand pass.
• Shell — coding-features status, curl -s localhost:3034/health.

Full Verification Guide

First usage¶

Start a session¶

cd ~/my-project
coding                     # best available agent
coding --claude            # or --copilot, --opencode, --pi
coding --project DIR       # another directory
coding --dry-run           # resolve everything, launch nothing

The learning repo¶

With a learning tier, the first launch in a repo asks where its learned data goes:

Enter creates a private <repo>-history (via gh), a teammate's URL clones and shares theirs, skip keeps the data local and untracked. Never asked again for that repo. See Per-Repo Tenancy.

Look at what was learned¶

• Dashboard — Observations, Digests and Insights tabs at localhost:3032.
• Knowledge viewer — the graph, filtered by your active teams: vkb (opens 127.0.0.1:12436/viewer/coding).

Update the knowledge base¶

Within a session, ask for it in chat:

"ukb"         # incremental analysis (recent changes)
"ukb full"    # the entire history
"ukb debug"   # single-stepping

Or from a shell: semantic workflow run wave-analysis --team coding. It runs in the background for 10–20 minutes; follow it on the dashboard.

Share with your team¶

coding sync               # what each learning repo has to push / pull
coding sync --push        # asks, then pushes

Teams & Shared Learning

Next steps¶