good-hypothesis — Claude Code skill for testing hypotheses
The good-hypothesis Claude Code skill helps assess research explanations and shape hypotheses that can be tested. It loads during Explain or Test when candidate explanations seem weak.
MIT ★ 93 updated 2026-11-15
npx skills add Insilico-org/plsi --skill good-hypothesis -a claude-codeWhat good-hypothesis does
The skill helps judge whether a hypothesis earns its place by ruling out a possible outcome. It sorts falsifiers by what could check them: existing data, a study that can be found, or a measurement nobody has made yet.
Candidates that predict the same result should be merged; an observation that separates the remaining candidates becomes the next question. Evidence without an origin counts against a candidate, not in its favor. Leave untouched candidates open. A measurement that has not been made is a research direction, not a test.
What it helps with
- write a falsifier that can be checked
- merge candidates with matching predictions
- separate a test from a measurement that does not exist yet
- account for evidence with no known origin
Example prompts for the agent
- Assess this explanation and name an observation that could disprove it.
- Compare these candidates: what observation would make their predictions differ?
- Review the evidence and leave untouched candidates open.
How to install good-hypothesis
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill good-hypothesis -a claude-code2Manually: put the files in place
~/.claude/skills/good-hypothesis/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/good-hypothesis/
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill good-hypothesis -a codex2Manually: put the files in place
.agents/skills/good-hypothesis/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill good-hypothesis -a cursor2Manually: put the files in place
.cursor/skills/good-hypothesis/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Insilico-org/plsi skills/lab/good-hypothesis2Manually: put the files in place
.github/skills/good-hypothesis/SKILL.md1Skills CLI (Vercel)
npx skills add Insilico-org/plsi --skill good-hypothesis -a gemini-cli2Platform CLI
gemini skills install https://github.com/Insilico-org/plsi --path skills/lab/good-hypothesisManually: put the files in place
.devin/skills/good-hypothesis/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.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 good-hypothesis
How do I install good-hypothesis in Claude Code?
Run npx skills add Insilico-org/plsi --skill good-hypothesis -a claude-code. The skill file is at ~/.claude/skills/good-hypothesis/SKILL.md.
What does good-hypothesis do?
It helps judge what a hypothesis rules out and identify an observation that could falsify it. It also guides merging candidates with matching predictions and keeping untested ones open.
When does good-hypothesis load?
It loads during the Explain or Test move in research when candidate explanations seem thin.
Source text
SKILL.md · MIT
--- name: good-hypothesis description: Use as PLSI Research Orchestrator when proposing or judging a candidate explanation in research - what makes one worth having, how to write a falsifier something can actually touch, when two candidates are really one, and when to kill or supersede. Load it from the Explain or Test move when the candidates feel thin. --- # What makes a hypothesis worth having The move pages say what to produce: `lab/explain` (the boring three first, `if_false`, `tested_by`, sharpening as a new node) and `lab/test` (personal data first, the context question, closing what died). This says how to tell a candidate that earns its place. **A hypothesis earns its place by what it rules out.** "Something about the exposure affects sleep" excludes nothing. "The effect is dose dependent" excludes a flat dose-response. ## The falsifier has to name something that could exist `save` refuses a candidate with no `if_false` but cannot judge it. Three grades, and only the first two are tests: | Grade | The falsifier is met by | What to do | |---|---|---| | **cheap** | something already in the graph, or in the user's own record | test it now, at no retrieval cost | | **reachable** | a study that exists and can be found | that is a Gather | | **wished for** | a measurement nobody has made | **not a test.** Record it as a direction, the question that would settle it, and say in the body that nothing can touch it yet | ## Two candidates that predict the same thing are one Ask of each pair: **what observation would come out differently between them?** If none, merge them and say so in `dropped`. The observation that separates the survivors is the next question. ## Unattributed evidence, untouched candidates - **Unattributed evidence counts against a candidate and not for it**: a finding with no origin may repeat one already counted. - **Leave a candidate open** when nothing has touched it. Many untested candidates and no evidence means Gather was needed.
© Insilico-org · MIT · Insilico-org/plsi