Documentation
Start
-
Getting started
Install addit-harness as a Claude Code plugin, run setup, and check that the subagents loaded.
-
Concepts
Why addit-harness exists, the engineering loop, deterministic orchestration with /dev-flow, tiered language conventions, and the curation philosophy.
-
The engineering loop
How the always-on engineering loop works, where every plan, design, report and brief is written, and the doc protocol that keeps those documents short and evidence-based.
-
A worked example: from idea to merged change
A start-to-finish walkthrough of one small change with addit-harness — install, setup, /addit-harness:dev-flow, the approval gate, where the documents land, and committing.
Reference
-
Subagents
The 15 subagents addit-harness ships, what each one does, which model it runs on, and which tools it is allowed.
-
How dev-flow works
How /dev-flow triages a request into light, standard or deep, runs design and review loops as Workflow scripts, holds a hook-enforced plan-approval gate, and verifies with QA once.
-
Skills & commands
Every slash command addit-harness adds, the Claude Code built-ins it reuses, and the official plugins it declares.
-
Model & cost
Which model each subagent runs on, how explicit tool lists and dev-flow tiers keep token use down, and manual levers to cut spend.
-
Enabling MCP
How to enable the optional MCP servers in mcp.example.json, from Figma and Atlassian to databases and cloud providers.
-
Local telemetry
The optional, off-by-default usage log addit-harness can keep on your own machine. What it records, what it never records, how to turn it on, and how to export a summary.
Guides
Project
Edit this page on GitHub ↗