Compatibility

This page tracks the conformance status of known CKP implementations. Each entry shows how many test vectors pass, skip, or fail, and the resulting verdict.

Results table

Implementation	Level Tested	Pass	Skip	Fail	Verdict
@clawkernel/sdk	L3	31	0	0	L3 CONFORMANT
ckp-bridge	L1	13	0	0	L1 CONFORMANT
ckp-python	L3	31	0	0	L3 CONFORMANT
NanoClaw Bridge	L3	31	0	0	L3 CONFORMANT
NullClaw Bridge	L3	31	0	0	L3 CONFORMANT
NanoClaw	L1	4	9	0	L1 PARTIAL
PicoClaw	L1	4	9	0	L1 PARTIAL
NanoBot	L1	4	9	0	L1 PARTIAL

---

Understanding the verdicts

CONFORMANT

All test vectors for the declared level pass with zero skips, zero failures, and zero errors.

PARTIAL

Some vectors pass but at least one is skipped. No vectors fail. This indicates that the implementation handles tested scenarios correctly but does not fully cover every requirement at that level.

NON-CONFORMANT

At least one vector fails — the implementation returned an incorrect response. This verdict means the runtime does not meet the requirements for that level.

---

Implementation details

@clawkernel/sdk

The official TypeScript SDK. Zero runtime dependencies. Supports all three conformance levels through a single createAgent() factory function.

• 31 vectors pass across L1, L2, and L3
• L1: 13/13 pass
• L2: 10/10 pass
• L3: 8/8 pass
• See SDK Reference for usage

ckp-bridge

A minimal reference implementation (~215 lines of TypeScript) that demonstrates the L1 wire protocol. It handles initialization, status, shutdown, heartbeat, and error conditions.

• 13 vectors pass (all L1)
• Full L1 CONFORMANT — zero skips, zero failures

ckp-python

An independent CKP implementation in pure Python (~350 lines) using only the standard library. Shares no code with the TypeScript SDK. Implements the full L1+L2+L3 wire protocol from the specification alone — proving CKP is truly language-agnostic.

• 31 vectors pass across L1, L2, and L3
• L1: 13/13 pass
• L2: 10/10 pass (tool pipeline, policy, sandbox, approval, quota)
• L3: 8/8 pass (memory store/query/compact, swarm delegate/report, broadcast/discover, manifest-validation negatives)
• Full L3 CONFORMANT — zero skips, zero failures
• Zero external dependencies — Python 3.9+ standard library only

NanoClaw Bridge

A CKP stdio bridge for NanoClaw that exposes JSON-RPC handlers for lifecycle, tools, memory, and swarm methods.

• 31 vectors pass across L1, L2, and L3
• L1: 13/13 pass
• L2: 10/10 pass
• L3: 8/8 pass
• See NanoClaw profile for architecture and evidence

NullClaw Bridge

A CKP stdio bridge for NullClaw — a Zig-based AI assistant runtime (~96K LOC, 678KB static binary) with 30+ tools, multi-sandbox auto-detection, hybrid memory, and leader-worker swarm coordination.

• 31 vectors pass across L1, L2, and L3
• L1: 13/13 pass
• L2: 10/10 pass
• L3: 8/8 pass
• Maps NullClaw’s Zig-native subsystems (sandbox, memory, swarm) to CKP JSON-RPC primitives

NanoClaw (native baseline)

An open-source WhatsApp-based AI agent framework. NanoClaw was not built for CKP but implements many of the same concepts (identity via CLAUDE.md, Docker sandboxing, per-group memory). These baseline results are manifest-only and skip transport vectors because native NanoClaw does not expose stdio JSON-RPC by default.

• 4 vectors pass (manifest validation vectors)
• 9 vectors skipped at L1 (no native JSON-RPC transport)
• See NanoClaw profile for full baseline and bridge evidence

PicoClaw (native baseline)

PicoClaw is currently assessed as a manifest-only baseline. It maps well to CKP concepts, but does not yet expose a native CKP JSON-RPC transport for live vector execution.

• 4 vectors pass (manifest validation vectors)
• 9 vectors skipped at L1 (no native JSON-RPC transport)
• See PicoClaw profile for baseline evidence and bridge next steps

NanoBot (native baseline)

NanoBot has strong MCP and orchestration primitives, but the audited runtime does not yet expose a native CKP stdio JSON-RPC surface.

• 4 vectors pass (manifest validation vectors)
• 9 vectors skipped at L1 (no native JSON-RPC transport)
• See NanoBot profile for baseline evidence and bridge next steps

---

How to add your implementation

Run the conformance harness against your agent and submit the results:

# Clone and build the harness
git clone https://github.com/angelgalvisc/ckp-test.git
cd ckp-test && npm install && npm run build

# Run against your agent
node dist/cli.js run \
  --target "node path/to/your/agent.js" \
  --manifest path/to/claw.yaml \
  --level 3

See Testing for complete instructions on running the harness and interpreting results.