What problem does it solve?
Deciding when and how to delegate work to subagents is error-prone: spawning workers without clear ownership wastes tokens, merges conflict, and unverified worker output pollutes results. This Skill provides a host-agnostic operating model for task decomposition, specialist delegation, coordination, synthesis, and local Ollama offload so parallel work stays bounded and verifiable.
Core Features & Use Cases
- Delegation gate and decomposition: Decides solo vs batch vs spawn, builds a dependency-aware DAG of worker tasks, and writes sealed packets with goal, scope, acceptance, and return shape.
- Coordination and synthesis: Portable list/wait/steer/stop actions, barrier-before-synthesize merging, conflict-first resolution, and recovery escalation for stalled or failing workers.
- Challenge techniques: Rubber-duck, interview, mimic-flow, red-team, blind review, and consensus patterns for challenging claims with fresh-context critics.
- Local Ollama offload: GATE → ROUTE → RUN → VERIFY → REPORT loop for tool-less one-shot or map-reduce jobs (summarize, extract, classify, translate, vision) using live
ollama list inventory and tier-based model selection.
- Use Case: You have 40 large log files to classify. The skill gates the decision, routes to a small installed Ollama model, runs sealed packets per shard via
scripts/ollama-worker.sh, verifies each return against a schema, and reports what was offloaded.
Quick Start
Ask the agent to break your large task into parallel workers with clear ownership, or to offload summarization of saved files to a local Ollama model after running the health check script.
