Skip to content
Canon

Volume V

The Harness

Skills, subagents, sandboxes, files

A harness is the opinionated runtime around the loop: progressive-disclosure skills, a filesystem workspace, isolated subagents, a sandbox, and a planner. Deep Agents, DeerFlow, OpenHands, and Claude Code all converge on this shape.

  1. V.1What a Harness IsA harness is the opinionated product around the loop: workspace, tools, skills, subagents, permissions, UI. Frameworks are libraries. Harnesses are how an engineer actually works with a model.7 min
  2. V.2Skills and Progressive DisclosureA skill is a folder with a SKILL.md. Metadata always sits in context; the body loads on trigger; scripts and references load on demand. This is how you scale expertise without flooding the window.8 min
  3. V.3SubagentsA subagent is a nested loop with a narrower prompt, a fresh or sliced context, and a written return value. It is the isolation primitive of modern harnesses.6 min
  4. V.4Sandboxes and IsolationEnforce boundaries at the tool and sandbox layer, not by asking the model to behave. E2B, OpenHands, and Deep Agents all say the same sentence.6 min
  5. V.5Planning and TodosLong-horizon work needs an external plan the model can read and rewrite. A todo list is a tool, not a thought.5 min
  6. V.6Filesystem as WorkspaceThe filesystem is the agent's memory, bus, and artifact store. Notes, plans, patches, screenshots — if it must survive a compaction, it is a file.6 min
  7. V.7Coding Agent HarnessesOpenHands, Cline, OpenCode, Aider, Gemini CLI, DeerFlow: the software-engineering specialization of the harness. Git is the ledger; the repo is the workspace; the PR is the done state.7 min