AGENT SKILL

hypothesis-design

Use when an idea is about to move from open exploration into empirical testing — sharpening it into falsifiable hypotheses and pre-registering what would change your mind before any run happens. Not for early free-form brainstorming; that belongs in ordinary chat.

License 许可
MIT
所属仓库
yibocat/prismnext
最近更新
2026-09-05
Stars
★ 4
Forks
0

01 前言 Front Matter

name
hypothesis-design
license
MIT
description
Use when an idea is about to move from open exploration into empirical testing — sharpening it into falsifiable hypotheses and pre-registering what would change your mind before any run happens. Not for early free-form brainstorming; that belongs in ordinary chat.

03 SKILL.md 正文

Hypothesis Design

The pre-registration gate between exploration and empirical work. Open-ended brainstorming and early pressure-testing happen in ordinary chat (Research design module). Invoke this skill when an idea is ripe enough that you are about to spend runs on it — decide what would change your mind before you see the data. This is the anti-HARKing discipline.

Be a sparring partner, not a judge: stress-test the clarity of a hypothesis, never veto the user's idea. The only hard no is retrofitting hypotheses to results already seen.

When to use

  • An idea is about to consume empirical effort (runs, data collection)
  • Sharpening a claim into falsifiable hypotheses with explicit kill conditions
  • Pre-registering claims, metrics, and stopping criteria before experiments

Workflow

  1. Read the spine — research-brief-read. Situate the idea in the existing through-line (or note that it diverges).
  2. Candidate hypotheses — one sentence each, falsifiable: state the observation that would kill it. Label every one as a candidate.
  3. Per hypothesis —
    • Evidence kinds that would support vs weaken it
    • Confounds and alternative explanations
    • The cheapest decisive test available in this workspace: existing data or artifacts on disk, a seconds-long probe script via experiment-run, or a small slice of an existing pipeline — check these before proposing any new full-scale experiment.
  4. Pre-commit — per hypothesis: metrics, thresholds, seeds, stopping criteria, and what each possible outcome changes next. Write this before running.
  5. Confirm, then persist — walk the user through the candidates; after confirmation:
    • Hypotheses and claims → research-brief-update (one section per call, first person).
    • Metrics, thresholds, seeds, stopping criteria → the experiment design artifacts (design matrix / island), not the brief — the brief holds the story, not the protocol.
  6. Hand off — empirical design → the design-matrix step (experiment-design-matrix when enabled); running → an island with experiment-log; multi-phase programs → suggest-plan.

Done when

  • Every hypothesis has a stated kill condition and is labeled a candidate.
  • Every hypothesis names its cheapest decisive in-workspace test.
  • Pre-commitments are written before any run, and the user has confirmed them.
  • Claims live in the brief; protocol details live in the design matrix / island.

Rules

  • Hypotheses stay labeled candidates until evidence lands.
  • Never retrofit hypotheses to observed results — that is HARKing; flag it plainly when it happens.
  • Present weaknesses and alternative explanations honestly, but the decision to pursue an idea is always the user's.
  • Orchestrator may delegate a challenge pass to research-design-coach; experts cannot Task — pressure-test inside your own deliverable.

04 同仓库更多 Skills

yibocat
yibocat/prismnext

demo-multi-file

Local test skill with an extra notes file

★ 4
yibocat
yibocat/prismnext

writing-results

Use when drafting a Results / Experiments / Analysis section — narrating verified evidence from run receipts, with complete statistical reporting, derivation analysis, and honest treatment of negative results. Every number traces to a run; every claim survives the audit.

★ 4
yibocat
yibocat/prismnext

writing-related-work

Use when drafting a Related Work / literature review section or a standalone synthesis note — organizing prior work into a grounded, cited narrative or comparing methods across papers, from the project library plus staged external discovery. Never cites from memory.

★ 4
yibocat
yibocat/prismnext

writing-preliminaries

Use when drafting a Preliminaries / Background / Notation section — definitions, problem setup, and formal statements the rest of the paper builds on, with notation consistency and machine-checked formal claims. Reference patterns, not molds; include only what later sections actually use.

★ 4
yibocat
yibocat/prismnext

writing-methods

Use when drafting a Methods / Methodology / Approach / Model section — expounding the ideas and the reasoning behind the method, grounded in the project's brief, notes, and prior discussions, with empirical details from run receipts. Structure follows the idea's shape; reference angles, not fixed templates.

★ 4
yibocat
yibocat/prismnext

writing-introduction

Use when drafting or rewriting a paper's Introduction — framing the problem, positioning against prior work, and stating contributions as checkable claims, grounded in the brief, notes, and the library. Reference rhetorical patterns, adapted to the paper — never applied as molds.

★ 4