OpenClaw SysML v2 Export

Exports your OpenClaw instance as a SysML v2 model, the OMG's standard language for describing systems: your agents and their models, the tools each one may use, their instructions, skills and memory, cron jobs and heartbeats, channels, sandboxes and exec settings. On request, each agent's own model also proposes the procedures and rules its instruction files describe.

Every run is kept as a version and shown in an offline page: an overview of the agents, the system, a tools allocation matrix, scheduled tasks over a week and the inferred behaviour. Click any element to read its SysML text; what changed since the previous run is highlighted and listed, value by value.

Status: preview. Requires OpenClaw 2026.10.9 or later.

Install

openclaw plugins install clawhub:@casys/openclaw-sysmlv2-export

Use

openclaw sysmlv2-export                 # export the instance
openclaw sysmlv2-export --infer --open  # also infer behaviour, then open the viewer

• In a chat, the owner can run /sysmlv2_export, or /sysmlv2_export infer to infer with the model of the conversation's agent.
• As an agent tool, sysmlv2_export exports without calling a model. The tool is optional: allow it explicitly, for instance with tools.alsoAllow.
• From the command line, --infer analyses each agent with its own model and needs plugins.entries.sysmlv2-export.llm.allowAgentIdOverride: true; --model <provider/model> also needs allowModelOverride: true.

Each run writes <state dir>/sysmlv2-export/<UTC time>/ with snapshot.json, findings.json, validation.json and the SysML files, and updates versions.json and the viewer, index.html.

Check the system

With the Toolkit installed, every export checks the SysML model with OpenClaw and APE SHACL rules. The summary counts findings and the viewer's Doctor panel lists them. If the Toolkit is unavailable, model checks are skipped and collector notes remain available. Findings do not block an export or change the instance.

See the diagnostics guide for all codes and their sources.

Validate the model

Each export can be compiled with sysml-toolkit, an open-source SysML v2 compiler. Install it once, explicitly; exporting never downloads anything.

openclaw sysmlv2-toolkit install [--from <archive.tgz>]
openclaw sysmlv2-toolkit status

install fetches the pinned public sysml-wasm 0.10.10 release (about 21 MB), or reads a local copy, and checks it against its pinned SHA-256. Every export is then compiled with a 60-second deadline. APE runs under its own default 10-second deadline; APE_TIMEOUT preserves the compile result. The result says whether validation passed, failed or was skipped, and validation.json lists the diagnostics. The Toolkit is under the Apache License 2.0 and its standard library under the EPL 2.0; their licenses are installed with it.

Privacy

• The export only reads the instance and never changes it.
• Secret values, tokens, API keys, URLs, people's ids and host paths are never exported. Files are referenced by path and digest, never copied.
• Inference sends only an agent's behaviour files (AGENTS.md, SOUL.md, TOOLS.md…) and its cron and heartbeat prompts to the model of one of your agents, never MEMORY.md, USER.md or other notes. Contact details and tokens are masked in the answer, and the proposals are kept apart, never admitted automatically.

Source, documentation and specification: https://github.com/Casys-AI/openclaw-sysmlv2-export. MIT license; the bundle includes elkjs 0.9.3 (Eclipse Public License 2.0) to lay the diagrams out, and THIRD_PARTY_NOTICES.md holds the notices of every bundled package.

Contributors: see contribution boundaries and the source map.