openclaw/Kova
openclaw/Kova
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
main
Kova/AGENTS.md
Shakker 1dabe239cd
feat: ship agent skills with Kova
2026-05-02 14:22:32 +01:00

2.4 KiB
Raw Permalink Blame History

AGENTS.md

Kova is an OpenClaw runtime validation lab.

Kova uses OCM as the environment/runtime control plane, but Kova is not an OCM test suite. Reports, scenario names, failures, and recommendations must stay focused on OpenClaw behavior.

Kova is agent-first and human-usable. Agents should prefer structured JSON outputs for planning and report consumption, while Markdown reports should stay short enough for humans to scan.

If the agent runtime has an ocm-operator skill available, use it before running real Kova scenarios. The skill teaches the agent how to use OCM safely: env cloning, runtime selection, local release builds, service status, logs, cleanup, and durable user-state boundaries. Kova remains OpenClaw-focused; the skill only improves the harness operations.

The repo-local OCM skill lives at .agents/skills/ocm-operator.

Core Rules

  • Treat OpenClaw as the product under test.
  • Treat OCM as harness infrastructure.
  • Use the ocm-operator skill when available before executing Kova scenarios that create envs, clone existing state, build local runtimes, upgrade envs, or inspect services/logs.
  • Prefer release-shaped OpenClaw runtimes over source/dev commands when judging release behavior.
  • Do not mutate durable user envs directly. Clone existing user envs before testing upgrade or migration scenarios.
  • Keep reports quiet: pass/fail/blocker, concise evidence, and actionable OpenClaw owner area.
  • Prefer node bin/kova.mjs plan --json when an agent needs to choose scenarios programmatically.
  • Prefer the JSON report for detailed evidence and the Markdown report for human summaries.
  • Do not parse verbose command stdout from Markdown. Successful command stdout belongs in JSON only.
  • Clean up temporary envs and artifacts created by executed scenarios unless the user explicitly asks to retain them.
  • If a scenario fails because Kova or OCM could not provision the lab, mark it BLOCKED. If OpenClaw starts but is slow, broken, missing files, leaking memory, stalling health, or failing plugins, mark it FAIL.

Command Style

Use the local CLI during development:

node bin/kova.mjs help
node bin/kova.mjs plan
node bin/kova.mjs plan --json
node bin/kova.mjs run --target npm:2026.4.27 --scenario fresh-install

kova run is safe by default and currently writes dry-run reports. Real execution must remain explicit and cleanup-aware.