OpenClaw OA Operator

Use this skill to operate OA as an observability and self-heal layer for OpenClaw. Focus on stable workflows, explain what the data means, and keep public packaging separate from machine-specific local setup.

Quick Start

• Inspect the workspace for config.yaml, self_heal_rules.yaml, scripts/, patches/, pipelines/, and any dashboard override such as dashboard-zh/.
• Determine whether the user wants local operations, dashboard refinement, or publishable packaging. Do not switch UI structure without explicit approval once the user picks a preferred layout.
• Prefer the simplest stable path: use upstream OA behavior unless the workspace already carries justified overrides.
• Run the smallest useful verification after each meaningful change: API check, script smoke check, or dashboard screenshot.
• Use scripts/oa_workspace_smoke_test.sh for a minimal end-to-end workspace check when OA is already configured.

Operate OA Locally

1. Confirm the current OA surface.

Use existing workspace scripts first if present, especially wrappers such as scripts/start_oa_server.sh, scripts/stop_oa_server.sh, and scripts/manage_oa_launchd.sh.

1. Verify the runtime before editing UI.

Check that oa collect can populate data and that oa serve exposes the expected endpoints before diagnosing the dashboard.

1. Treat configuration files as policy.

Read config.yaml for goals, thresholds, OpenClaw path, and agent scope. Read self_heal_rules.yaml before changing remediation behavior.

1. Keep verification concrete.

Validate /api/health, /api/goals, /api/team-health, /api/traces, and any self-heal endpoints used by the current dashboard.

Interpret Metrics And Self-Heal State

• Explain G1 and G2 first because they are usually the most stable built-ins: cron reliability and team health.
• Explain custom goals only after confirming their pipelines and metric sources exist in the workspace.
• Distinguish clearly between:
• signal quality issues
• pipeline failures
• dashboard rendering issues
• self-heal policy breaches
• For self-heal work, preserve the loop:
• detect
• fix
• verify
• learn
• Before running command-mode fixers, inspect the concrete command template, placeholders, and target paths. Prefer non-destructive fixers and keep fallback ticketing enabled when available.

Refine Dashboards Safely

• Preserve the user's chosen layout once they say a version is clearer. Improve copy, density, and data mapping before changing structure.
• Prefer upstream OA layout when the user values clarity and comparability to the original repository.
• Use project-level overrides only when there is a clear need for localization, extra tabs, or richer diagnostics.
• After UI edits, verify both desktop and mobile views with real screenshots instead of assuming the layout works.

Package For ClawHub

• Prefer a skill when the value is workflow knowledge: install OA, operate it, interpret metrics, and guide self-heal decisions.
• Prefer a plugin when the value is a long-running service, HTTP routes, background workers, or a productized dashboard experience.
• Remove machine-specific defaults before public packaging:
• absolute paths
• local ports and launchd labels
• local agent rosters
• user-specific ticket directories
• local backup and artifact paths
• Package reusable instructions, checklists, and safe defaults. Do not package private workspace snapshots as public examples.
• Read references/release-readiness.md before preparing a public release.
• Read references/smoke-test.md before claiming the package is publish-ready.

Deliverables

• For local operations, report what changed, how it was verified, and what remains risky.
• For release preparation, produce:
• a public versus private diff
• a publish-readiness checklist
• proposed listing copy
• a recommendation of skill versus plugin