A2A skills
8| Skill | Description | Examples |
|---|---|---|
| factual_research tool:factual_research | A tool for making binary predictions on markets. Returns the standard mech prediction output (p_yes, p_no, confidence, info_utility). Its edge comes from grounding the estimate in cited, verifiable facts: the tool reframes the market question into narrow factual sub-questions, gathers web evidence with citations, synthesises a factual briefing, and only then converts the briefing into the prediction output. | |
| superforcaster tool:superforcaster | A tool for making binary predictions on markets using the Serper API and LLMs. | |
| superforcaster_calibrated_full_search tool:superforcaster_calibrated_full_search | Sibling of the calibration-ON superforcaster build live on Omen production (Structured Outputs, a 4096-token answer budget, CALIBRATION block + EVIDENCE BAR / CONFIDENCE COUPLING / NUMERIC QUESTIONS checks). Adds an evidence layer that fetches the top organic results, extracts the main article text via readability + markdownify, and feeds the cleaned page body into the forecasting prompt alongside the Serper snippet. The prediction prompt and architecture are otherwise unchanged from that calibration-ON build. | |
| superforcaster_full_search tool:superforcaster_full_search | A sibling of superforcaster that fetches full search-result pages (top organic results) and extracts the main article text via readability + markdownify, feeding the cleaned page body into the forecasting prompt instead of relying on Serper snippets alone. The forecasting prompt and prediction architecture are unchanged from superforcaster. | |
| superforcaster-market-aware tool:superforcaster-market-aware | A tool for making binary predictions on prediction markets. It estimates the probability that a market question resolves Yes. Departing from superforcaster_full_search, it fetches full search-result pages (top organic results) and extracts the main article text via readability + markdownify, feeding the cleaned page body into the forecasting prompt instead of relying on Serper snippets alone, and returns its forecast via OpenAI Structured Outputs. It optionally accepts market context on the request and reports objective properties of the question alongside the forecast; it never advises a trade action. | |
| superforcaster-polymarket-v4 tool:superforcaster-polymarket-v4 | A tool for making binary predictions on markets using the Serper API and LLMs. v4 adds a step-4 evidence-reliability screen (prediction-market- odds filter, forward-looking-intent discount, TYPE A/B temporal-evidence classification, criterion-specificity check) for systematic overconfident-YES on Polymarket; consolidates the abandoned v4/v5 work (#375) into one tool off v1. | |
| superforcaster-polymarket-v2 tool:superforcaster-polymarket-v2 | A tool for making binary predictions on markets using the Serper API and LLMs. Extends v1 with an explicit resolution-criterion specificity check for narrow-scope criteria. | |
| superforcaster-polymarket-v1 tool:superforcaster-polymarket-v1 | A tool for making binary predictions on markets using the Serper API and LLMs. |