{"id":19207,"plugin_id":"plugins_6a92f40359bc81919e5b41b1451ec501","kind":"skill","collection_source":null,"comparison_source":null,"observed_at":"2026-09-30T23:15:23.527Z","digest":"81543c9a479ec472baea8f2dd92fcc477e5298ed0a7c31ce8d40e52b39cd7933","against":null,"payload":{"description":"Use when coordinating multiple AI forecasting perspectives for Power BI sales forecasts, including backlog, seasonality, budget, sales skepticism, risk, consensus, dissent, and explainable forecast arbitration.","included_files":[{"relative_path":"agents/openai.yaml","size_in_bytes":148}],"name":"powerbi-autonomous-forecast-agents","skill_md_contents":"---\r\nname: powerbi-autonomous-forecast-agents\r\ndescription: Use when coordinating multiple AI forecasting perspectives for Power BI sales forecasts, including backlog, seasonality, budget, sales skepticism, risk, consensus, dissent, and explainable forecast arbitration.\r\n---\r\n\r\n# Power BI Autonomous Forecast Agents\r\n\r\nUse this skill when the user wants an agent council rather than one blended forecast. The goal is a structured debate that exposes why the forecast is high, low, risky, or trustworthy.\r\n\r\n## Agent roles\r\n\r\n- **Backlog Agent**: trusts open orders, delivery dates, status, order age, and learned conversion.\r\n- **Demand Agent**: trusts seasonality, trend, intermittent demand, and customer/product history.\r\n- **Budget Defense Agent**: explains the gap to budget and roll forecast.\r\n- **Sales Skeptic Agent**: searches for optimism bias, volatile customers, sparse products, and weak evidence.\r\n- **Risk Agent**: flags data quality, missing snapshots, memory fallback, biased backtests, delivery risk, and low-confidence matching.\r\n- **Arbitrator Agent**: creates the final consensus, dissent, confidence, and recommended action.\r\n\r\n## Workflow\r\n\r\n1. Load the latest `ai-forecast-summary.csv`, `ai-forecast-detail.csv`, `ai-forecast-top-deltas.csv`, and `ai-forecast-model-quality.csv` when available.\r\n2. Run each agent independently against the same month and grain.\r\n3. Require every agent to provide:\r\n   - forecast direction: up, down, or hold\r\n   - evidence\r\n   - confidence\r\n   - top 3 risks\r\n   - recommended action\r\n4. The arbitrator produces:\r\n   - consensus forecast\r\n   - dissenting views\r\n   - final risk flag\r\n   - whether AI forecast can replace, challenge, or only annotate the roll forecast\r\n\r\n## Output columns\r\n\r\n- `forecast_month`\r\n- `grain`\r\n- `customer`\r\n- `product`\r\n- `agent_name`\r\n- `agent_forecast`\r\n- `agent_confidence`\r\n- `evidence`\r\n- `risk_flag`\r\n- `recommended_action`\r\n- `arbitrated_decision`\r\n\r\n## Guardrails\r\n\r\n- If model quality is worse than roll forecast, mark the result `advisory_only`.\r\n- Do not hide dissent. Dissent is the value of the council.\r\n- Use concrete customer/product/month evidence, not generic commentary.\r\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}