Decision Orchestrator Skill

# Decision Orchestrator Skill English | [简体中文](./README.zh-CN.md) A production-oriented skill that runs a full **Router -> Advisors -> CEO Final** decision workflow in one command. Designed for: - Codex workflows - Claude Code workflows - OpenClaw pipeline handoff ## 1. Why This Skill Manual multi-prompt orchestration is slow and inconsistent. This skill turns it into a deterministic run package: - Standardized structure - Traceable decision artifacts - Faster execution from idea to action - OpenClaw-compatible payload output ## 2. Workflow Architecture ```text Input JSON | v Router (classify + route) | v Advisor Chain (domain-specific analysis) | v CEO Final (conflict merge + final decision + action plan) | v Run Artifacts + OpenClaw Payload ``` ### Routing Logic (default) - Growth Strategy: `Drucker -> Buffett -> Musk` - Product/Experience: `Jobs -> KenyaHara -> Musk` - High-Impact Decision: `Munger -> Drucker -> Buffett -> Musk` - Execution-Stuck: `Musk` ## 3. Advisor/Prompt Responsibilities ### Router Prompt - Goal: classify case type, route advisors, identify data gaps. - Output: category/confidence/route/handoff notes. - Constraint: no final business decision. ### Drucker (Value Definition) - Focus: customer identity and value clarity. - Checks: customer-fit, value proposition, evidence quality. - Hard stop: no clear customer value. ### Jobs (Experience Quality) - Focus: user flow clarity and decision friction. - Checks: flow coherence, message clarity, content reduction. - Hard stop: user cannot understand value quickly. ### Kenya Hara (System Simplicity) - Focus: structural necessity and cognitive load. - Checks: remove/merge/rename for minimal complexity. - Hard stop: complexity increases without clarity gain. ### Munger (Multi-Model Decision) - Focus: cross-model risk decomposition. - Checks: economics + psychology + probability consistency. - Hard stop: key assumptions untestable or tail risk unmanaged. ### Buffett (Moat and Focus) - Focus: strategic moat reinforcement. - Checks: moat gain, imitation risk, opportunity cost. - Hard stop: clear strategic drift from core positioning. ### Musk (Execution Velocity) - Focus: move from analysis to immediate execution. - Checks: 48-hour launchable action plan, measurable criteria. - Hard stop: no owner or no measurable deliverable. ### CEO Final Prompt - Goal: merge conflicting advisor outputs and finalize decision. - Output: final decision, tradeoffs, 48h/7d actions, kill criteria. - Constraint: explicit accountability and review date. ## 4. Run Artifacts Each run folder includes: - `01-decision-input.md` - `02-router-output.yaml` - `03-agent-outputs.yaml` - `04-ceo-final.yaml` - `05-decision-log.md` - `06-review-summary.md` - `07-one-shot-prompt.md` - `openclaw-payload.json` ## 5. Quick Start ```bash python3 scripts/run_multi_advisor.py \ --input references/openclaw-input.example.json \ --runs-root ./runs \ --owner "Decision Owner" ``` ## 6. OpenClaw Integration Use the same input schema as `references/openclaw-input.example.json`. Output `openclaw-payload.json` can be passed directly to downstream nodes for: - artifact indexing - post-processing - notification/reporting ## 7. UI Profile Config (Not GitHub Account Identity) This section controls only **skill UI display metadata** inside `agents/openai*.yaml`. It does **not** control your GitHub commit account name/avatar. - Default profile: `agents/openai.yaml` - Alternate profile: `agents/openai.chatgpt.yaml` To switch profile style, copy the alternate file content into `agents/openai.yaml` and commit. ## 8. Repository Structure ```text decision-orchestrator-skill/ ├── SKILL.md ├── README.md ├── README.zh-CN.md ├── agents/ │ ├── openai.yaml │ └── openai.chatgpt.yaml ├── assets/ ├── references/ │ └── openclaw-input.example.json ├── docs/ │ └── agents.md └── scripts/ └── run_multi_advisor.py ``` ## 9. Versioning Guidance - Use semantic tags for workflow/runtime changes. - Keep prompt-structure changes backward-compatible where possible. - Log routing or scoring logic changes in commit messages. ## 10. License Use and adapt freely for your own workflow repository. ## 11. Detailed Prompt Spec For a strict output contract by role, see [docs/agents.md](./docs/agents.md).