“The refund tool ran.”
McPherson Governance for OpenClaw
See what your OpenClaw agents did before you give governance the power to stop them.
McPherson Governance evaluates supported tool activity against explicit policy and records what happened, what policy would have decided, what needs review, and what remains unknown.
The current release operates in shadow mode. It has no authority to block, approve, deny, or rewrite your agents’ actions.
Why activity logs are not enough
A tool event and governance evidence answer different questions.
“Who requested it, which policy applied, whether approval was required, whether the action completed, and whether the evidence is sufficient.”
Illustrative example only. It does not describe a specific production integration or claim that the current release observes every refund workflow.
How it works
Build evidence without overstating authority.
Shadow mode is designed to evaluate policy without actively controlling production actions.
- Agent proposesA supported tool action enters the configured path.
- Plugin observesAn attempt receipt and minimized metadata are produced.
- Policy evaluatesExplicit rules produce a non-authoritative shadow decision.
- Evidence is boundDirectly observed completion is correlated to the attempt.
- Observa explainsMeasured facts stay separate from assumptions.
- Operator reviewsGaps and exceptions become follow-up work.
Agent proposes. Policy decides. Runtime enforces. Observa explains.
In the current v0.5 release, policy decisions are recorded in shadow mode. Runtime enforcement is not active.
Current release
McPherson Governance v0.5.1
Public install contract
- Corrected installation and lifecycle documentation
- Named-profile state isolation
- Available through ClawHub
Runtime guardrails
- OpenClaw 2026.6.5 or newer required
- Runtime compatibility gating
- Disabled mode is inert for ordinary observations
Verified release
- v0.5.1 passed 250 release checks with 0 failures and 0 skipped.
- Public-install smoke test passed
- Authority remains shadow-only
Coverage and limitations
Observation has a boundary. The product should say where it is.
Coverage depends on configuration
- Coverage depends on supported hooks, correct mappings, selected tools, and the configured observation path.
- Actions outside that path may not be represented.
- A missing receipt does not prove that no action occurred.
Shadow decisions do not execute
- A shadow policy decision does not control production execution.
- McPherson Governance does not certify compliance, safety, or completeness.
- It has no authority to block, approve, deny, or rewrite your agents’ actions.
Proof
Operator proof and technical proof serve different readers.
What an operator needs to review
- What was attempted
- What completed
- What policy would have decided
- What needs review
- What remains unknown
What the public release can substantiate
- McPherson AI release verification for v0.5.1
- 250 release checks with 0 failures and 0 skipped
- Release checksum and source link
- Shadow-authority verification
- Documented privacy boundary
- ClawHub availability, clean scan status, and isolated install smoke test
Free plugin and managed setup
Choose the level of help your environment needs.
For technical operators
For technical OpenClaw operators who can install, configure, map, and validate their own environment.
The public v0.5.1 install-contract patch is live. It corrected installation and lifecycle documentation, added named-profile state isolation and runtime compatibility gating, and keeps disabled mode inert for ordinary observations.
Install the Free PluginFor operators who want a guided start
For operators who want McPherson AI to help map tools, define policy, review shadow evidence, and identify coverage gaps.
Request Founding SetupRoadmap
Observation first. Authority only after gates are verified.
v0.5
Observe supported activity and produce shadow policy decisions and evidence.
v0.6
Discover capabilities, assist mappings, detect drift, and diagnose governability.
v0.7
Introduce bounded enforcement only after policy, runtime, approval, and evidence gates are verified.
FAQ
Questions to resolve before relying on the connector.
Does the current release block my agents?
No. The current release evaluates policy in shadow mode. Remote decisions have no execution authority, and the connector does not actively block, approve, deny, delay, or rewrite ordinary tool actions.
What data leaves my environment?
When enabled, the connector can send a closed allowlist of identifiers, class labels, one-way hashes, timestamps, and mapped metadata to the governance endpoint you configure. It does not send prompts, message bodies, tool parameters, file contents, credentials, customer records, or local receipt ledgers. Operator-chosen identifiers can themselves be sensitive, so naming and retention still require review.
What happens if the policy service is unavailable?
The connector remains non-blocking. A slow, malformed, absent, or hostile remote response can change the local observation record, but it receives no authority over the ordinary production action. Bounded queues and circuit behavior also mean some observations may be dropped and recorded as such.
Does it observe every tool?
No. Observation depends on supported OpenClaw hooks, correct manual mappings, selected tools, and the configured path. Unmapped activity may produce an unknown-tool observation, and activity outside the path may not be represented.
Can I use it without managed setup?
Yes. The free public plugin is intended for technical OpenClaw operators who can verify the package, configure the endpoint and private state paths, map tools, start disabled, and validate the evidence themselves. Review the current release notes before installing.
Does it certify compliance?
No. McPherson Governance does not certify compliance, security, safety, visibility, or completeness. Its evidence can support review, but the meaning and sufficiency of that evidence remain bounded by coverage and operator context.
What happens before enforcement receives authority?
Policy behavior, runtime integration, approval handling, evidence quality, failure behavior, and operator controls must be verified as explicit gates. The roadmap does not grant the current release authority it does not have.
How do I install or uninstall it?
Use the tagged public installation guide: verify the package and file manifest, install through OpenClaw’s supported mechanism, configure it disabled, check status, and enable deliberately. For removal, disable first, unpair if needed, run node <install-dir>/connector/connector-ctl.mjs uninstall, and reload the OpenClaw gateway. The tagged lifecycle guide preserves receipts by default and documents irreversible data removal separately.
What OpenClaw versions are supported?
The public v0.5.1 release requires OpenClaw 2026.6.5 or newer and refuses incompatible runtimes. That is a compatibility floor, not a guarantee for every later host version or environment.
McPherson Governance is an independent product and is not affiliated with or endorsed by the OpenClaw project.