harness-kit Wiki
harness-kit is a set of Claude Code agents that take an idea to a merged PR through one gated pipeline, plus a system-architect agent that writes System Design Docs. The repo holds the terse agent-facing files. This wiki holds the theory and the study material. Start with Harness Engineering.
The harness
| Page | Covers |
|---|---|
| Getting Started | install, pick a setup, and go from a brief to a merged PR, step by step |
| Harness Engineering | Agent = Model + Harness, feedforward and feedback, humans on the loop |
| References | the sources behind each design choice |
| Guides | templates, examples, writing style, conventions |
| Sensors | deterministic structure checks and hard gates |
| Evals | atomic-check rubrics, local or jev judge, pass at 8.0 |
| Pipeline and Stages | the six stages, markers, tokens, context tools, SDD loop |
| Graph Engineering | requirement ids, trace manifest, scope gate, FalkorDB graph |
| Graph Theory | requirement atoms, traceability, code property graphs, temporal knowledge graphs |
| Jev and System One | LLM judge bias, calibrated probabilities, atomic checks, escalation |
| Golden Path | /golden-path, its five properties, detours |
| Agents | product-manager, staff-software-engineer, system-architect |
| Agent Pipelines | the stage flow of each agent |
| Designer Skill | M3, dark and light theme, i18n, favicon for new UIs |
Autonomy and subagents
You approve direction at two gates and everything between runs on its own. An intake stage harvests context first, and an orchestrator dispatches single-purpose subagents.
| Page | Covers |
|---|---|
| Autonomy | intake, resolve-mark-proceed, /pipeline:run |
| Orchestration and Subagents | orchestrator and leaves, writer and critic split, model tiering |
The canonical shape is .claude/shared/pipeline-pattern.md. Haiku tiering and per-org repos.md are still (planned).
System design
The system-architect agent runs one playbook per episode of the System Design series. A design doc is an optional front stage before the main pipeline: it feeds a sharper PRP and plan.
| Page | Design problem |
|---|---|
| System Design Method | the 13-stage method, three pillars, back-of-envelope numbers |
| URL Shortener (#1) | read-heavy lookup, short-code generation, cache, abuse |
| Rate Limiter (#2) | token bucket, fail-open, multi-region budgets |
| Search Engine (#3) | crawler, indexing, ranking, query serving |
Next candidates are API gateway, distributed queue, news feed, distributed cache, and notification system.
Global references
Birgitta Böckeler, Harness engineering for coding agent users and Maintainability sensors for coding agents, martinfowler.com, 2026. Martin Kleppmann, Designing Data-Intensive Applications, O'Reilly, 2017. Jeff Dean, Designs, Lessons and Advice from Building Large Distributed Systems, LADIS, 2009. Werner Vogels on design for failure, Pat Helland on immutability, Michael Nygard's Release It!, John Ousterhout's A Philosophy of Software Design. Eval research is on References.
This page mirrors harness-kit Wiki in the wiki. Edit it there.