This practice guide covers isolation, guardrail tools, and credential hygiene for OpenClaw. One layer it doesn't yet reference: a dispatch-layer enforcement boundary that produces a tamper-evident audit trail of what tool calls actually executed (vs. what was requested).
We're building HELM AI Kernel (OSS, MIT), a fail-closed execution firewall that sits between the OpenClaw agent and its MCP servers:
- Intercepts all tool calls via an OpenAI-compatible proxy
- Enforces default-deny policy per tool call (ALLOW/DENY/ESCALATE)
- Quarantines unknown MCP servers until schema and policy approval
- Emits a signed receipt per dispatch decision
- Bundles receipts into an offline-verifiable EvidencePack (
helm-ai-kernel verify evidence-pack.tar)
This closes a gap the guide currently leaves open: isolation controls (VMs, containers) prove the environment is restricted, but they don't produce cryptographic proof of which tool calls ran within that environment during a specific session.
Suggested guide addition
Under "Recommended Setup for Experimentation", consider adding:
HELM AI Kernel — OSS fail-closed dispatch boundary; default-deny at the MCP proxy layer; produces signed receipts and offline-verifiable EvidencePacks per session
https://github.com/Mindburn-Labs/helm-ai-kernel
Local demo (~5 min):
brew install mindburnlabs/tap/helm-ai-kernel
helm-ai-kernel serve --policy ./release.high_risk.v3.toml
helm-ai-kernel boundary status --json
helm-ai-kernel receipts tail --agent agent.demo.exec --server http://127.0.0.1:7714
helm-ai-kernel verify evidence-pack.tar
One feedback question: Does this practice guide currently recommend any tool that produces a post-session verifiable record of which tool calls executed — or is that gap acknowledged somewhere?
This practice guide covers isolation, guardrail tools, and credential hygiene for OpenClaw. One layer it doesn't yet reference: a dispatch-layer enforcement boundary that produces a tamper-evident audit trail of what tool calls actually executed (vs. what was requested).
We're building HELM AI Kernel (OSS, MIT), a fail-closed execution firewall that sits between the OpenClaw agent and its MCP servers:
helm-ai-kernel verify evidence-pack.tar)This closes a gap the guide currently leaves open: isolation controls (VMs, containers) prove the environment is restricted, but they don't produce cryptographic proof of which tool calls ran within that environment during a specific session.
Suggested guide addition
Under "Recommended Setup for Experimentation", consider adding:
Local demo (~5 min):
One feedback question: Does this practice guide currently recommend any tool that produces a post-session verifiable record of which tool calls executed — or is that gap acknowledged somewhere?