what-if-oracle — Claude Code skill: scenario planning for research
Walks a research plan through what-if branches: favorable, reference, adverse and more. Shows where the plan breaks before you spend resources on it.
npx skills add K-Dense-AI/scientific-agent-skills --skill what-if-oracle -a claude-codeWhat what-if-oracle does
The skill runs in three phases. You write one operational question with a decision, a comparator, a horizon and success criteria, then keep a short evidence ledger: claim, source or data version, and status — observed, model-derived or assumed. Next you pick four to six branches: favorable, reference, adverse, wild card, contrarian and second order. Each branch gets a compact record covering condition, assumptions, mechanism, consequences, a discriminating observation, a trigger and residual risk.
A branch label is an exploration lens, not a probability. The skill states plainly that scenario narratives do not estimate causal effects and do not produce calibrated forecast probabilities. When a question needs a power calculation or a numerical simulation, it tells you to name the required design and evidence instead of writing prose around the gap.
No packages, API keys or network access are needed for the qualitative workflow. External facts that matter to the decision you verify yourself, or you label the analysis as built only on supplied material. Runs in Claude, Codex, Cursor, GitHub Copilot, Gemini CLI and OpenCode.
What it helps with
- lay out a research plan across several possible futures
- prepare a fallback for lower-than-planned sample yield
- check which option stays feasible under resource constraints
- separate observations from assumptions in an evidence ledger
- set triggers and thresholds that force a plan review
Example prompts for the agent
- Break our sampling plan into what-if branches: what changes if the usable fraction comes in below plan
- Build an evidence ledger for this protocol and mark what is observed versus assumed
- Run a contrarian branch against our main hypothesis and tell me which observation would weaken it
How to install what-if-oracle
1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill what-if-oracle -a claude-code2Manually: put the files in place
~/.claude/skills/what-if-oracle/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/what-if-oracle/
1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill what-if-oracle -a codex2Manually: put the files in place
.agents/skills/what-if-oracle/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill what-if-oracle -a cursor2Manually: put the files in place
.cursor/skills/what-if-oracle/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install K-Dense-AI/scientific-agent-skills what-if-oracle2Manually: put the files in place
.github/skills/what-if-oracle/SKILL.md1Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill what-if-oracle -a gemini-cli2Platform CLI
gemini skills install https://github.com/K-Dense-AI/scientific-agent-skills --path skills/what-if-oracleManually: put the files in place
.devin/skills/what-if-oracle/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add K-Dense-AI/scientific-agent-skills --skill what-if-oracle -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about what-if-oracle
How do I install what-if-oracle in Claude Code?
Run npx skills add K-Dense-AI/scientific-agent-skills --skill what-if-oracle -a claude-code. The skill file then sits at ~/.claude/skills/what-if-oracle/SKILL.md.
What does what-if-oracle do?
It runs structured scenario analysis for research planning and scientific project decisions: branches, assumptions, evidence and decision triggers.
Do I need API keys or packages?
No. The qualitative workflow needs no packages, credentials or network. Verify external facts that affect the decision separately.
