Skip to content

chronulusai/chronulus

v0.2.8MIT

Teaches your agent to use the Chronulus MCP server (https://mcp.chronulus.com/mcp): binary outcome and multi-outcome prediction with BinaryPredictor and time-series forecasts with the NormalizedForecaster.

chronulus

Skills for Claude Code, Claude.ai and OpenAI Codex that teach your agent how to use the Chronulus MCP server. This plugin includes two skills: predict and forecast.

The plugin also works on Claude.ai and ChatGPT.com, and the MCP server includes UIs built specifically for those platforms (such as the prediction and forecast scorecards).

predict

A skill that teaches Claude how to use the chronulus-mcp Chronulus MCP server to build reusable, calibrated probability predictors for prediction-market questions (Kalshi, Polymarket, and similar) — binary markets like "will A beat B?", "will X happen by date D?" or "will Y exceed a threshold?", and markets with more than two outcomes, like a soccer win/draw/loss line, an ordinal ladder (a rate-decision market), or an exclusive-winner field (an award category, a tournament, a multi-candidate race).

Category references

skills/predict/references/ has one guide per broad market category, each with the market shapes, a session template, input fields, an evidence checklist and the pitfalls specific to that category:

CategoryReferenceDetailed subcategory guides
Sportssports/index.mdfootball, soccer, tennis, volleyball, basketball, e-sports (CS2, League of Legends, Dota 2, Valorant)
Electionselections/index.mdplanned
Politics and policypolitics/index.mdplanned
Culture and entertainmentculture/index.mdplanned
Cryptocrypto/index.mdplanned
Commoditiescommodities/index.mdplanned
Climate and weatherclimate/index.mdplanned
Economicseconomics/index.mdplanned
Mentionsmentions/index.mdplanned
Financefinance/index.mdplanned
Tech and sciencetech-science/index.mdplanned

What it covers

  • Setting up one session + one agent per (category, bet type) that gets reused across every future question, instead of recreating either per question.
  • Two input schemas: a matchup schema (event-level fields plus symmetric side1_*/side2_* fields) for contests, and a proposition/negation schema for thresholds, "by date" events, draws and multi-outcome markets.
  • Data-collection guidance: research instead of fabricate, explicit "not available" notes over blanks/guesses, and keeping evidence parallel between sides.
  • The dual-framing technique — run every question twice with the framing swapped, then average the Beta parameters — to cancel directional framing bias before trusting or quoting a probability.
  • Batching: running many predictions concurrently in one call with batch_reuse_prediction_agents_and_get_predictions — both framings of a question, a slate of games, or every outcome of a multi-outcome market — including the 32-expert batch cap and per-item error handling.
  • Beyond binary: eliciting one dual-framed, debiased Beta per outcome (via a proposition/negation input schema instead of fixed sides) and reconciling them into a single Dirichlet with reconcile_ordinal_outcome_dirichlet / reconcile_exclusive_outcome_dirichlet and their render_*_dirichlet_scorecard counterparts — covering ordinal outcomes (a draw between two win outcomes, a "hold" between cuts and hikes) as well as unordered exclusive-winner fields.
  • Comparing against market prices: unit conversion (Kalshi cents, Polymarket dollars), tradable prices net of fees, and keeping market prices out of the inputs.

forecast

A skill for producing time-series forecasts with the Chronulus NormalizedForecaster: no historical data required, a 0–1 normalized series plus the agent's explanation, over a horizon of hours, days or weeks.

  • Setting up one session + one agent per forecasting use case and reusing it across items that share an input_data_model.
  • Writing the session situation/task, designing the input fields, and the same research-don't-fabricate data-collection rules.
  • Setting forecast_start_dt_str, time_scale and horizon_len.
  • Showing the result with render_forecast_scorecard (plot plus explanation tooltip) instead of hand-building a chart.
  • Rescaling into real units with y_min/y_max and invert_scale, including the exact formula, and when to use rescale_forecast for the numbers themselves.
  • The optional, on-demand risk assessment scorecard.

Requirements

A Chronulus account. This plugin bundles a connection to the hosted Chronulus MCP server at https://mcp.chronulus.com/mcp (the production deployment of chronulus-mcp) — no local install or API key file needed. You'll be prompted to sign in via OAuth the first time a tool from it is used.

Installation

Claude Code

Install via this marketplace:

/plugin marketplace add ChronulusAI/plugin-marketplace
/plugin install chronulus

Run /mcp afterward to confirm the Chronulus server connected and to complete sign-in.

OpenAI Codex

Add the marketplace, then install chronulus from the Plugins directory in Codex:

codex plugin marketplace add ChronulusAI/plugin-marketplace