Which AI coding agents does Harnd work with?

Any of them. `harnd scan` writes a plain-Markdown brief into `.harnd/agents/` that Claude Code, Cursor, Copilot, Codex — or whatever you switch to next — can read. And `harnd validate` runs in your terminal and in CI, so the gate holds no matter which agent wrote the code.

Is my source code sent anywhere?

harnd init, scan, and validate are fully local — no account, no network, nothing leaves your machine. Only sync talks to the cloud, and only if you opt in. It sends harness metadata (token and component names, gates, usage counts, decisions), the file paths those live at, and the single offending line of each violation, up to 200 characters. Harnd never uploads your codebase, and sync --redact omits even that line.

What do I need to run Harnd?

Node.js 20 or later and a Git repository. No design tool, and no account for the local commands.

What does Harnd cost?

Free while in beta. Waitlist members get first access to pricing and an early-partner offer before anyone else.

← All docs

Configuration

harnd init writes .harnd/config.yaml, and it is safe to hand-edit. Here is what every key in it means.

.harnd/config.yaml is generated by harnd init and updated by a handful of other commands (link, rename, refresh). The file itself says it is safe to hand-edit — it is the source of truth, not a cache of one.

schema_version: "2.0.0"
harness:
  id: hrn_01k...
  name: my-project
  description: ""
project:
  root: ..
defaults:
  gates:
    - no-raw-color
    - no-raw-spacing
scan:
  components:
    - "packages/ui/src/**"
  paused:
    reason: "waiting on a known scan bug"
    until_version: "0.3.0"
cloud:
  linked: false
  api_url: https://harnd.com
  harness_id: hrn_...

schema_version

Required. A string in "2.x.y" form. harnd refresh updates an existing .harnd/ to the schema the installed CLI expects; it is a dry run by default, and applies changes only with --write.

language

Optional. "en" or "ja", default "en". This is the language of generated files — for example .harnd/agents/harness.md and the rationale text on default gates. It is not the language of the CLI's own output, which is always English regardless of this setting. Set it with harnd init --lang ja.

harness

harness.id — Generated once at init (hrn_…), and never changes after that, even across init --force.
harness.name — A human-readable name. The recommended way to change it later is harnd rename <name>, not hand-editing this field directly — though the file as a whole is safe to hand-edit.
harness.description — Optional, free text.

project.root

The path from .harnd/ to the repository root, relative. In almost every setup this is .., since .harnd/ lives one level below the root.

defaults.gates

Internal bookkeeping: the list of default gate rules this harness has already been offered. harnd refresh uses it to tell a gate you deliberately removed apart from a gate that did not exist yet when you ran init — without it, refresh could not tell the two apart and would resurrect gates you removed on purpose. It exists so removed gates never come back; it is not meant for hand-editing.

scan.components

Optional. An array of repo-relative glob patterns declaring where to look for components. By default, scan only looks in directories whose purpose is unambiguous by name — components/, ui/, and similar conventions, plus route-colocated files under app/. Set this explicitly for a monorepo (packages/ui/src/**) or a non-standard layout, such as Feature-Sliced Design's src/features/**/ui/**.

This is detection scope, not validation scope. Gates have their own, separate scope.include. Conflating the two can wrongly mark real components as deprecated when they fall outside a gate's scope but are in fact still in use.

scan.paused

Optional. Declares that you have deliberately stopped running scan. A paused: block is not for hiding that you stopped — the staleness warning keeps showing either way. It changes the warning from “you're being negligent” to “declared, with a reason.”

scan.paused.reason — Required if paused: is present at all. A paused: block with no reason is treated as malformed and ignored — harnd says so explicitly rather than silently accepting it.
scan.paused.until_version — Optional. Use this when the reason is “waiting on a specific harnd bug or version” rather than a standing policy. Once you are running that version or later, scan and validate will tell you that you can probably clear the pause. Harnd never clears it automatically — you always do that by hand.

cloud

All of the local commands — init, scan, validate — leave this section alone. It only changes once you run harnd link.

cloud.linked — Set to true by harnd link. Always false right after init, since local commands need no account.
cloud.api_url — Recorded by harnd link (default https://harnd.com). Not secret.
cloud.harness_id — The cloud-side id, recorded if known. Informational only: harnd sync resolves the actual destination from your API key, not from this field.