Openclaw universal system

OpenClaw Universal System

Purpose

This skill makes OpenClaw a machine-global capability instead of a repository-local convention. It applies across all VS Code Copilot agent chats on this machine, regardless of the current folder or project.

Global Scope

The canonical shared assets live here:

• Strategy: ~/.copilot/openclaw/OPENCLAW_UNIVERSAL_SYSTEM.md
• Quick reference: ~/.copilot/openclaw/OPENCLAW_UNIVERSAL_QUICK_REF.md
• Routing map: ~/.copilot/openclaw/task-router-profile-map.json
• Bootstrap command: ~/.local/bin/openclaw-bootstrap-repo

Repository docs may add overlays, but they should not redefine the global policy.

To connect a repository to this system, run:

• global-skills-bootstrap-repo /absolute/path/to/repo init
• global-skills-bootstrap-repo /absolute/path/to/repo audit

openclaw-bootstrap-repo remains available as a compatibility wrapper and delegates to the higher-level global-skills bootstrap.

The bootstrap keeps OpenClaw organized within the wider global-skills stack by updating or creating the repository's global-skills.instructions.md contract and then maintaining a dedicated OpenClaw overlay file.

Activation Rule

Load this skill whenever a task involves one or more of:

• multi-file implementation
• architecture analysis or planning
• heavy testing or browser automation
• performance or static analysis
• long sessions where compute pressure may grow

For tiny edits, local execution is acceptable unless another repository policy overrides it.

Universal Control Loop

Use this sequence in every project:

CLASSIFY -> PRECHECK -> ROUTE -> EXECUTE -> OBSERVE -> VERIFY -> AUDIT

1. CLASSIFY
   • Map the task to a universal task type using the routing map.
2. PRECHECK
   • Reuse openclaw-availability-utilization for gateway health and utilization checks.
3. ROUTE
   • Prefer OpenClaw for complex, long-running, high-memory, or multi-step work.
4. EXECUTE
   • Run the selected slice on the chosen lane.
5. OBSERVE
   • Emit or preserve decision evidence: task type, rationale, model/profile, confidence, lane.
6. VERIFY
   • Run repo-local tests or validation gates.
7. AUDIT
   • Compare actual usage against the target pattern and adapt the next slice.

Universal Checkpoints

For substantive implementation work, evaluate OpenClaw at three checkpoints:

• Design: planning, decomposition, trade-off analysis
• Implementation: heavy coding, test generation, large refactors
• Verification: broad test suites, regression checks, release validation

Each checkpoint should produce a short rationale:

• why OpenClaw was used, or
• why local execution remained sufficient

Task Routing Defaults

Use the shared routing map to classify work into:

• code-generation
• analysis
• reasoning-advanced
• documentation
• testing

Default guidance:

• code-generation, analysis, reasoning-advanced, testing -> prefer OpenClaw when scope is medium or heavy
• documentation -> local is acceptable unless the session already benefits from OpenClaw continuity

Observability Minimum

Every project should capture at least this lightweight evidence in progress updates or logs:

• lane used
• task type
• confidence or complexity estimate
• rationale summary
• verification outcome

Integration

Load together with:

• operator-identity-context
• repo-standards-router
• openclaw-availability-utilization
• network-platform-resources when remote/offloaded execution matters
• workflow-self-anneal after repeated drift or failures

Success Standard

Global success means:

• the policy is reusable across repositories
• canonical guidance is stored under ~/.copilot
• repository docs reference the global system instead of acting as the only source
• future projects can adopt the same operating model without copying the TSV Ledger docs