Execute spec-driven builds through feedback loop

conductor-orchestratorskillsetup L3★0

What it does

Dispatch spec-driven tasks to right skill-cluster via closed-loop resolution

Best for

Multi-organ builds where each task needs the right capability loadout (not generic) and visual work needs brand taste injected before code generation.

Inputs

· tasks.md (checkbox queue)
· plan.md (technical context + routing stack)
· spec.md (appetite/no-go)

Outputs

· resolved {cluster, dispatch, tier, activate, spokes, confidence} per task
· subagent loaded with <cluster>-orchestrator
· ship-battery.mjs verdict (fail-closed gates)

Requires

· resolve-task.mjs (classifier+validator)
· skill-clusters/skill-index.json
· conducty (loop phases)
· taste-resolve.mjs (visual clusters)

Preconditions

skill-index.json populated; resolver trained on real clusters; modality (local|github-delivery) chosen upfront

Failure modes

· phantom cluster (not in skill-index) → hard error
· low-confidence task → escalate to human (don't dispatch blind)
· taste brief omitted for visual cluster → shipped without brand alignment

Trust signals

· resolve-task contract documented
· Taste-learning feedback loop (taste-feedback.mjs records cycle)
· PAI fail-closed gates prevent non-enumerated skills