OpenClaw Constitutional Governance Skill
The first published reference implementation of constitutional AI governance for an open-source agentic platform — demonstrating that the AOS framework is vendor-agnostic and integrates with real-world agent infrastructure.
Published: February 1, 2026 · Repository: aos-openclaw-constitutional · Commits: 15
Why This Matters
A governance framework only matters if it works with real-world AI infrastructure — not just in theory. The OpenClaw Constitutional Governance Skill proves three things:
AOS governance is not locked to a single AI provider. The skill works with OpenClaw, which can be paired with any language model — OpenAI, Anthropic, Google, DeepSeek, or open-source models. The governance layer sits above the model layer.
This is not a whitepaper or concept document. It is working code — a constitution spec, cryptographic signing tools, deterministic classifiers, and policy evaluators — published in a public GitHub repository with 15 verifiable commits.
The AOS Constitutional Governance Skill was published on February 1, 2026 — before OpenClaw's creator announced joining OpenAI on February 15, 2026. This timeline is independently verifiable through Git commit timestamps.
What the Skill Contains
The repository provides a complete Phase 1 reference implementation:
Constitution Specification
- constitution.yaml — Human-readable constitutional policy with the 10 AOS bedrock amendments
- constitution.c14n.json — Deterministically canonicalized JSON for cryptographic signing
- constitution.attestation.json — GitTruth attestation contract example
Cryptographic Tooling
- c14n.py — Canonicalization script (YAML → deterministic JSON)
- sign.py — Ed25519 digital signature generation
- verify.py — Ed25519 signature verification
- verify_gittruth_stub.py — GitTruth attestation validation
Deterministic Policy Engine
- risk.py — Deterministic risk classification for agent actions
- classify.py — Deterministic tag classification against constitutional rules
- evaluate.py — Full policy evaluation engine (DENY / CONFIRM / ALLOW)
Important Note
This is Phase 1: artifacts, specification, verification, and reference algorithms. The README explicitly states: "Real immutability requires Gateway/tool-router enforcement (Phase 2). A Skill alone cannot prevent an agent from attempting prohibited tool calls." This honest limitation is part of the evidence record.
How to Verify
Every claim on this page is independently verifiable:
1. Verify the Repository Exists
2. Verify the Publication Date
The first commit (6e9c639) is timestamped February 1, 2026.
3. Verify the Constitution Hash
Compare against the signed attestation to verify integrity.
4. Run the Policy Engine
See the deterministic DENY/CONFIRM/ALLOW engine working.
Commit Timeline
All 15 commits occurred on February 1–2, 2026. Key milestones:
6e9c639Initial Release
AOS Constitutional Governance for OpenClaw — built via AI-to-AI collaboration
b3025f6Phase 1 Complete
Added classify.py and evaluate.py — full deterministic policy evaluation
a88c5e5Humanitarian Protections
40 prohibited categories: military, weapons, surveillance, violence, child exploitation, trafficking
c6296a1AOS Humanitarian License
Replaced MIT with irrevocable humanitarian license — military use prohibited permanently
e1ed21eFamily Consensus
All Family fixes applied: Oracle (legal), Arnold (patent), Scout (risk), Proto (technical)
c125f28v1.0.1 Clarity Release
Family-approved clarity improvements, 3.5 hours post-launch
What This Proves
AOS governance is not an "AOS-only" technology. It integrates with third-party agentic platforms, demonstrating that constitutional governance is a layer — not a replacement.
OpenClaw works with any language model. The AOS governance skill governs the actions, not the model — proving the architecture works regardless of the underlying AI.
Published February 1, 2026 — two weeks before the OpenClaw acquisition announcement. Git timestamps are immutable and independently verifiable.
The entire reference implementation is publicly available under the AOS Humanitarian License. The code, the constitution, the signing tools — all open for inspection.