What does this skill do, and when should you use it?
This is a pure prompt-based skill for making objective probability judgments about real-world future events. It enforces a reviewable workflow: clarify the Yes/No settlement definition first, design searches per evidence category, collect dated and linked evidence, assign direction and weight, build a structured conditional-probability model, then run a Bayesian-style update to produce a single headline probability with an 80% subjective confidence interval. It is market-blind by default: prediction-market prices are only recorded as an exclusion unless you explicitly ask to include them. Every output includes a model breakdown table, a key-evidence table, and an archive path for later review.
- Turns fuzzy questions into explicit Yes/No trigger conditions, deadlines, time zones, and official resolution criteria, flagging misjudgment-prone edges
- Decomposes the event into 2-5 necessary-condition nodes and designs search queries per node covering officials, mainstream media, party media, and third-party analysis
- Collects evidence with date, source, URL, excerpt, and reliability rating, distinguishing "talks ongoing" from "a settleable deal is reached"
- Weights each piece of evidence as strong/medium/weak support or opposition and models P(Yes) = P(A) × P(B|A) × ... from the settlement definition
- Runs a Bayesian-style update and outputs one headline probability, an 80% confidence interval, up/down factors, and Yes/No watch points
- An analyst needs a concrete probability that a regulatory proposal passes by year-end, with reasons and evidence attached
- A researcher wants a source-traceable estimate of the probability of a conflict escalation, ceasefire, or treaty signing
- A tech watcher wants the odds that a company ships a product or model before a given date
- A user wants a fair probability independent of prediction-market consensus, with market prices explicitly excluded
- An election follower wants a structured conditional breakdown of multi-condition events, like a candidate winning primary and then the general
- Situations requiring order placement or trade execution — this skill only produces probability research and calls no trading endpoints (that is a sibling skill in the same repo)
- Offline environments — time-sensitive questions must use web search, so the workflow cannot complete without network access
How do you install this skill?
- The skill depends on external web-search; the repo discloses the search layer once silently returned '0 results' with no evidence — verify search provider configuration before use.
- Probability outputs embed subjective weight judgments; this is an analysis framework, not fact, and should not be the sole basis for investment or trading decisions (the parent repo involves prediction-market trading).
- The skill files carry no version or changelog; cross-check commit history when pulling updates.
- Time-sensitive questions mandate live web search; mainland-China reachability of search sources is unverified.
- Network access
web-search capable tool for time-sensitive questions
This skill ships in the predict-raven repo (MIT licensed). The repo is driven through an AI agent (Claude Code / Codex / OpenClaw) in natural language; no standalone manual install commands are documented.
Claude Code
git clone https://github.com/Alchemist-X/predict-raven.git
cd predict-ravenStart your agent inside the cloned directory and say "install the dependencies for predict-raven"; the agent runs pnpm install + pnpm build (the collection's dependencies — the probability-analysis skill itself has none beyond a search-capable agent).
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- What is the fair probability that the Fed cuts rates before the December meeting? Give your reasoning and evidence.
- Analyze the probability of a Russia-Ukraine ceasefire within six months using your evidence framework; do not use Polymarket prices.
- What are the odds a new frontier model officially launches before October 1? Break it into conditional probabilities with key evidence.
- Will this bill pass in the current congressional session? Give me the headline probability, confidence interval, and Yes/No watch points.
Trigger it by asking a probability question: the chance of something happening, a win rate, a fair probability, or whether an edge exists. The skill then runs a seven-step flow: clarify definitions → design searches and gather evidence → list evidence → assign weights → build the conditional model → Bayesian update → output the headline probability with breakdown, evidence, and archive paths. Options: say "ignore prediction-market prices" to record them only as exclusions, or "include market signals" to allow market data. Paywalled sources keep only URL, title, date, and a short excerpt.
What are this skill's strengths and limitations?
- Forces a concrete probability with a confidence interval and archived evidence, making outputs reviewable after the fact
- Market-blind by default, so results are not just restatements of prediction-market prices
- Settlement-definition-driven conditional decomposition prevents arbitrary modeling
- Pure prompt skill with no scripts — any agent with web search can run it
- Outputs contain subjective elements: the 80% confidence interval and evidence weights are model self-assessments with no external calibration or Brier scoring
- Strongly dependent on web-search quality for time-sensitive questions; unreliable sources distort conclusions
- Both the evidence weights and the conditional probabilities come from the same model, with no independent verification
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Probability Analysis — Event Probability Analyst this page | 60 · Recommended | ★ 76 | 26d ago | MIT |
| Exa Deep Research Orchestrator | 57 · Use with care | ★ 5.1k | 18d ago | MIT |
| hv-analysis: Deep Research Report | 46 · Use with care | ★ 21k | 24d ago | MIT |
| Exa Agent Research Skill | 56 · Use with care | ★ 5.1k | 18d ago | MIT |
| TAO AutoML Tuning ✓ NVIDIA · Official | 54 · Use with care | ★ 3.6k | 11d ago | Apache-2.0 |
How did FollowSkills review this skill?
Pure prompt framework: no scripts, no permission requests, no network side effects, no sensitive-data handling; explicitly forbids fabricating sources and limits paywalled content to short excerpts; data flow transparent. Deducted for unverified publisher, no verifiable source attribution or rollback discussion, and static inability to confirm downstream agent compliance.
EN/ZH workflows are consistent and internally coherent with reproducible steps; but instruction text only — no skill-specific tests, thin failure-feedback design, and dependency on external web-search which the repo itself records as having silently failed (DDG CAPTCHA). Deducted for weak failure modes and unverified key paths (static cap 10).
Trigger conditions are explicit (probability, fair probability, edge), scenarios, exclusions and output format are all specified, with native Chinese version. Deducted for unspecified mainland-China reachability of mandatory web-search sources and only partial boundary definition for evidence-poor cases.
Clear bilingual layering, progressive disclosure, concrete prohibitions; MIT license at repo level. Deducted for no per-skill version, changelog, FAQ or worked examples, and no maintenance/update path stated inside the skill files themselves.
Structured conditional decomposition, evidence weighting and Bayesian update templates with a directly usable output format give marginal value over unstructured prompting. Deducted because output correctness is unverified statically, subjective weights lack external validity evidence, and the repo discloses a known search-quality defect that contaminated past outputs (static cap 7).
The skill mandates URL, date, excerpt and manifest paths per evidence item — built-in traceability. Deducted for no reproducible tests or third-party execution evidence for this skill; the repo's Brier track records target other product lines, leaving a single evidence type (static cap 5).
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →