soft-predict-future

What it does

Activates for ANY future-oriented question and requires web search to be enabled. It triggers on questions about future outcomes and on phrases like "predict" or "forecast". It enforces a 9-step pipeline: validate input, collect signals via six web searches, score signals, extract structural drivers, build a STEEEP matrix, perform cross-impact analysis, find three historical analogues, compute probabilities and confidence, and write scenarios plus assemble a report.

How it works

The skill prescribes a rigid, multi-step procedure:

• Step 1 — Validate Input: apply five binary rules to determine if the question is real, observable, time-bound, has evidence, and is specific enough; then output a PROCEED/STOP verdict with an inferred horizon if proceeding.
• Step 2 — Collect Signals: run exactly six web searches across categories (current state, growth data, barriers, policy, technology, historical analogue). For each signal, classify six attributes (direction, steeep_category, temporal_layer, source_type, recency_days, has_evidence) and present as a table.
• Step 3 — Score Signals: compute a base score using recency, reliability, type, and evidence multipliers, then apply a regional multiplier by STEEEP and temporal layer; cap at 1.0 and round to two decimals; provide a scoring table.
• Step 4 — Extract Structural Drivers: group signals by STEEEP and identify top 3 drivers with names, force statements, lists of explaining signals, temporal reach, and stability.
• Step 5 — Build 6×3 STEEEP Matrix: fill all cells with average final_scores, apply regional multipliers, and indicate hot zones, gaps, and dominant zone.
• Step 6 — Cross-Impact Analysis: assess convergence across temporal layers and identify friction points; compute a convergence bonus if strategic convergence exists.
• Step 7 — Find 3 Historical Analogues: select three analogues with similarity, tipping event, equivalence today, and driver validation.
• Step 8 — Compute Probabilities + Confidence: compute probable, plausible, possible scores with formulas, then convert to percentages; compute confidence from several contributing factors.
• Step 9 — Write Scenarios + Assemble Report: produce probable/plausible/possible/preferrable narratives, each traceable to a driver, with proof numbers/dates, activation conditions, constraints, and a DRIVER mapping; include a canonical output template.

When to use it

Use when the user asks any future-oriented question or uses keywords like predict, forecast, scenario analysis, or asks about races, adoption, geopolitics, or market dominance. If web search is unavailable, it will tell the user before proceeding.

What it can touch

The skill requires web search and uses the tool clause: claude-code. It references executing searches, evaluating signals, and compiling a structured forecast report.

Caveats

License: MIT. It explicitly states to activate only when the question meets the criteria and to proceed only with web-enabled data; otherwise, it informs the user. It uses a complex, multi-step pipeline with numerical scoring and probabilistic outputs; results depend on the quality and recency of signals collected from web sources.