Multitask Orchestrator

Multitask Orchestrator

Use this skill when one request contains multiple independent lanes that should run in parallel.

Trigger Conditions

Use multitask-orchestrator when all are true:

1. At least two lanes can proceed without blocking each other.
2. Each lane has a clear success artifact or pass/fail signal.
3. The final response must merge outcomes from all lanes.

Do not use this skill for single-lane work or tightly coupled sequential changes.

Workflow

1. Define lanes with stable IDs (lane-a, lane-b, ...), owners, and expected outputs.
2. Assign each lane a minimal chain (skill-hub routed) and explicit guardrails.
3. Run lanes in parallel.
4. Capture per-lane evidence (artifacts, checks, unresolved risks).
5. Merge lanes only after all required gates pass.
6. If any lane blocks, route that lane through request-loopback-resume and continue non-blocked lanes.

Lane Contract

Each lane must provide:

• objective: one sentence scope
• inputs: files/repos/skills used
• checks: commands or validation gates
• artifacts: JSON/log/doc paths
• status: pass, fail, or blocked
• next_action: deterministic follow-up

Merge Rules

1. Merge only pass lanes.
2. blocked lanes require loopback state before completion.
3. fail lanes must include root cause and remediation lane.
4. Final response includes:
   • lane summary table
   • unresolved risks
   • exact artifact references

Guardrails

• Keep lane scope minimal; avoid cross-lane file overlap when possible.
• Prefer deterministic scripts/evidence over free-form narrative.
• Healthy local OpenClaw-compatible subagents are preferred when a lane needs bounded sidecar analysis.
• If cloud sidecars are required, keep them lower-reasoning and low-cost.
• Do not suppress failing lane results to unblock completion.

References

• references/lane-contract.md