AGENT SKILL

hypothesis-review

Rank open hypotheses by testability and impact. Use when a research campaign needs to prioritize which hypothesis-driven experiment to run next.

License 许可
Apache-2.0
所属仓库
harmoniqs/amicode
最近更新
2026-11-12
Stars
★ 7
Forks
2

01 前言 Front Matter

name
hypothesis-review
agents
[ "researcher" ]
surface
public
description
Rank open hypotheses by testability and impact. Use when a research campaign needs to prioritize which hypothesis-driven experiment to run next.
project_contract
{ "folders": [ "ledger/hypotheses" ], "frontmatter": [ "status" ] }

03 SKILL.md 正文

Hypothesis Review

Rank open hypotheses to identify the highest-value experiments.

Process

  1. Locate hypotheses. Read from <project>/ledger/hypotheses/. Filter for notes with status: open or status: untested.
  2. For each hypothesis, read the body and assess:
    • Impact if confirmed (numeric): high=3 (unlocks fidelity breakthrough or new platform), medium=2 (improves existing result), low=1 (incremental or niche)
    • Ease of test (numeric): easy=3 (1 cold-start experiment), moderate=2 (needs specific setup or 2-3 experiments), hard=1 (requires engineering or new infrastructure)
    • Priority score = impact × ease (range 1–9)
  3. Sort by priority score descending
  4. Return top 3 hypotheses with: title, score breakdown, brief rationale for ranking, suggested experiment approach

Output Format

hypotheses:
  - path: "ledger/hypotheses/hypothesis-YYYYMMDD-HHMMSS-topic.md"
    title: "Free-phase may help fluxonium Y gate"
    impact: 3  # high — could unlock new-best on stuck gate
    ease: 3    # easy — 1 cold-start with free_phase=true
    score: 9
    rationale: "Y gate has 3/5 stagnation failures; free-phase proven on other platforms but never tried here"
    suggested_approach: "Cold-start with free_phase=true, cubic 21 knots, same T_init as current best"
  - ...

Edge Cases

  • If no hypotheses with status: open or status: untested exist, return empty list with note: "No open hypotheses to review."
  • If all hypotheses require engineering (ease=hard), note this and return them anyway — the researcher may still choose one.

04 同仓库更多 Skills