Skip to content
Papercup — Harness Docs

Papercup harness documentation

A supervised loop that runs autonomous coding agents — and tells you why it made every decision.
How the harness works Compared to other frameworks

What lives in these docs

Section titled “What lives in these docs”

These docs describe two layers of the Papercup system, both essential, neither sufficient on its own:

  1. Coding harness — the inner loop. A small bash supervisor that runs role agents (planner, worker, validator, orchestrator, etc.) on a single project. This is what physically writes code. One coding harness per project.
  2. Org layer — the outer loop. Five fixed director agents (Business, R&D, Technology, Management, Marketing) coordinating via typed inbox/outbox messages, dispatching directives into projects, each project running its own coding harness. This is the company-as-software.

Sections:

  • Overview — what Papercup is, how the coding harness works in 5 minutes, why we built our own.
  • Org layer — the company-level concepts (charter, directives, projects, messages, director runtime, full API + UI map).
  • Architecture — the role graph, memory model, checkpoints, worktrees, competition mode, smoke gates, decisions timeline. Coding-harness internals.
  • Decisions — the actual decision log: why fresh contexts per role, why a prompt-cache-dominant design, why a curator role.
  • Compared to other frameworks — head-to-head with gstack, Superpowers, GSD, Agentwise, Hermes, and seven others.
  • Operating — config reference, lifecycle hooks, glossary.