HWW — How We Work

Monkey Run's open methodology for multi-agent startup operations.

HWW is the operating standard that governs how Monkey Run builds, coordinates, and ships products using AI agents. It covers agent roles, file-based coordination, quality gates, execution patterns, and governance.

If Paperclip is the control plane, HWW is the playbook that runs on top of it.

What's Here

• docs/HWW-1.5.md — Current published standard
• docs/HWW-1.6-DRAFT.md — Draft: context engineering, discuss phase, goal-backward verification, state persistence
• docs/CONVENTIONS.md — Multi-agent operating conventions (entity resolution, PR decision tree, commit standards)
• research/ — Frameworks and tools evaluated against HWW
• templates/ — Reusable company templates (Paperclip-compatible)
• paperclip-eval/ — Live evaluation of Paperclip against Halo project

Current Version: HWW-1.5

Published. Deployed across: Halo, Commish Command, Backstage, Open Brain, monkey-run-baseline.

In Development: HWW-1.6

Four new sections inspired by GSD Framework analysis:

• §5 Context Management — fresh subagents per task, aggressive atomicity, wave parallelism
• §6 Discuss Phase — pre-sprint human input collection
• §7 Goal-Backward Verification — "what must be TRUE?" not "did we complete tasks?"
• §8 State Persistence — resume anywhere after session restart

Paperclip Evaluation

We're evaluating Paperclip as a control plane layer on top of HWW. Pilot project: Halo (investment dashboard). See paperclip-eval/ for findings.

Community

This methodology is open. If you're building with AI agents, steal what works.

MIT License.