Status: DRAFT — under review by the maintainers of OpenClaw, NanoClaw, and ZeroClaw. Nothing here is agreed yet.
Conformance profiles for agentic harnesses ("Claws"). This repo pins versions and subsets of three existing standards and defines how a harness proves conformance. It does not define protocols; protocol gaps go upstream.
| Standard | Layer | Governed by |
|---|---|---|
| Agent Skills | packaged knowledge | Anthropic (open spec) |
| ACP — Agent Client Protocol | client/UI ↔ agent | Zed Industries (Apache 2.0) |
| A2A — Agent2Agent | agent ↔ agent | Linux Foundation (Apache 2.0) |
RATIONALE.md— why these three standards, why now.spec/cip-1.0-draft.md— CIP-1.0 draft: conformance units, requirements, per-harness gap analysis, work items.
This is a proposal, not a standard. It was drafted by one of the three maintainers (ZeroClaw) and binds nobody:
- Draft (now): redlines via issue or PR from anyone; corrections from the OpenClaw and NanoClaw maintainers supersede any claim made here about their codebases.
- Accepted as CIP-1.0: only with explicit sign-off from all three maintainers. Sign-off is scoped — a maintainer's approval is required for exactly (a) normative text that binds their harness and (b) any published claim naming it. Nothing additive ever binds a harness whose maintainer didn't approve it.
- Exit: any maintainer can withdraw at any time; withdrawal removes every claim about their harness from this repo.
Redline anything: open an issue or PR against either document. The per-harness gap tables are outside readings of each codebase (SHA-pinned, evidence-linked) — corrections from maintainers supersede them.
The repo name and org are provisional. Moving to a neutral org, or renaming, is on the table — see RATIONALE §7.
Prose (RATIONALE.md, spec/): CC-BY-4.0. Schemas and test code: Apache-2.0.