Kova Operator

Mission

Use Kova to test OpenClaw as users actually run it. Kova is not a unit-test runner and not an OCM test suite. OCM is the lab control plane; OpenClaw is the product under test.

Kova should produce evidence a maintainer can act on:

• startup time, listening time, health readiness
• gateway, CLI, package-manager, runtime-staging, plugin, browser, and provider CPU/RSS attribution
• plugin load and missing dependency errors
• dashboard/TUI/API responsiveness
• agent turn breakdown: pre-provider OpenClaw time, provider time, post-provider time, cold/warm delta
• provider failure containment: timeout, malformed response, streaming stall, recovery, leaked children, gateway health
• repeat-run stats, baseline comparisons, and release gate verdicts

Do not overstate causality. If OpenClaw did not emit enough diagnostic spans, say Kova proved the outside-in timing but internal attribution is missing.

First Steps

1. Find Kova:

command -v kova || true
test -f bin/kova.mjs && node bin/kova.mjs version

Use kova when installed. Inside a Kova checkout, use:

node bin/kova.mjs <command>

1. Ensure OCM is available before real execution:

command -v ocm
ocm env list
ocm runtime list

If the ocm-operator skill is available, use it before running Kova scenarios that build runtimes, clone existing user envs, inspect services/logs, or clean up envs.

1. Set up Kova:

kova setup
kova self-check

For scripts/CI:

kova setup --ci --json
kova self-check --json

If using the repo checkout, replace kova with node bin/kova.mjs.

Install OCM Precisely

Kova real execution requires OCM. If OCM is missing, install it before running benchmarks:

command -v ocm || curl -fsSL https://raw.githubusercontent.com/shakkernerd/ocm/main/install.sh | bash
export PATH="$HOME/.local/bin:$PATH"
ocm --version
ocm env list
ocm runtime list

If installing Kova from its installer, OCM can be installed or required there:

curl -fsSL https://raw.githubusercontent.com/shakkernerd/Kova/main/install.sh | bash -s -- --install-ocm --require-ocm

For Codex agents working from this repo, use the repo-local OCM operating skill at .agents/skills/ocm-operator when direct OCM operations are needed.

Safety Rules

• Prefer mock auth first. It isolates OpenClaw latency from provider latency.
• Run live auth only as a separate lane and mark it environment-dependent.
• Prefer local-build:<repo> for release-like validation of local OpenClaw source.
• Do not use source/dev commands as proof a published package will work.
• Do not mutate durable user envs directly. Use Kova clone/source flows.
• Use --execute only when a real run is intended.
• Keep failing envs only when inspection is needed; otherwise let Kova clean up.
• Report OpenClaw failures as OpenClaw failures, not OCM failures, unless lab provisioning itself blocked the run.

OpenClaw Repo Handling

When the user asks to benchmark OpenClaw, first decide which OpenClaw source is under test:

1. Use the explicit repo path if the user gives one.
2. If the current directory is an OpenClaw checkout, use it.
3. If a nearby checkout exists, such as ../openclaw, use it only after confirming the path and commit.
4. If no checkout exists, clone a disposable copy into /tmp, not inside Kova:

repo="/tmp/kova-openclaw-$(date +%Y%m%d%H%M%S)"
git clone https://github.com/openclaw/openclaw.git "$repo"
git -C "$repo" checkout main
git -C "$repo" pull --ff-only

Before benchmarking a checkout, record:

git -C "$repo" status --short
git -C "$repo" rev-parse --abbrev-ref HEAD
git -C "$repo" rev-parse --short HEAD
git -C "$repo" log -1 --format='%h %ci %s'

Do not mutate a user working tree unless the user asked for that. For benchmarking moving targets, prefer disposable clones or worktrees under /tmp/kova-openclaw-*.

Use release-shaped Kova targets for local source:

…