OpenClaw Integration
Get your OpenClaw agent running on TraceCore's deterministic harness in under 5 minutes.
No OpenClaw install required to try it out.
Quick Start
1. Install
pip install -e ".[dev]"
2. Try it with the mock workspace (no OpenClaw needed)
cd examples/mock_openclaw_workspace agent-bench openclaw --agent-id log-monitor
This scaffolds tracecore_adapters/log-monitor_adapter_agent.py. Open it — the act() stub is ready to fill in.
3. Implement act() — AI IDE shortcut
If you're in Windsurf, Cursor, or another AI IDE, paste this prompt:
"Read tracecore_adapters/log-monitor_adapter_agent.py, workspace/AGENTS.md, and cron/jobs.json. Implement act() so the agent passes log_alert_triage@1. Run agent-bench openclaw --agent-id log-monitor --task log_alert_triage@1 --seed 0 and fix failures until it passes."
Available Commands
Detect and scaffold adapters
agent-bench openclaw --agent-id <id> agent-bench openclaw --gateway
Detects an OpenClaw agent from openclaw.json (CWD or ~/.openclaw/), scaffolds a self-contained TraceCore adapter, and runs it against a task. Auto-detects the agent ID when only one named agent exists.
Use --gateway to additionally scaffold a gateway-wired adapter that calls the OpenClaw gateway RPC (agent / agent.wait) per step.
Export certified bundles
agent-bench openclaw-export --agent-id <id>
Writes a certified bundle (<id>_adapter_agent.py, <id>_gateway_adapter_agent.py, AGENTS.md, openclaw.json, manifest.json, README.md) to tracecore_export/<id>/. Blocked until a passing run exists for the adapter.
Bundle adapters are for optional regression testing — not deployment; the OpenClaw agent continues to run normally in OpenClaw.
What Gets Generated
The scaffold creates a TraceCore-compatible adapter that:
- Implements the reset/observe/act interface
- Bridges OpenClaw's tool APIs to TraceCore actions
- Preserves the original agent's logic and state management
- Can be tested locally without OpenClaw runtime
Task Mapping
Common OpenClaw agent patterns map to TraceCore tasks:
- Log monitoring agents → log_alert_triage@1
- API interaction agents → rate_limited_api@1
- Multi-step workflows → rate_limited_chain@1
- Config management → config_drift_remediation@1
Next Steps
- Read the full tutorial at tutorials/openclaw_quickstart.md
- Check the mock workspace in examples/mock_openclaw_workspace/
- Review adapter patterns in the generated code
- Run baseline comparisons to validate behavior