Skip to content

momo-land/PHD-management-skills / skills/phd-experiment

phd-experiment

Verified

为科研项目设计可检验的实验、登记真实运行与结果、维护基线和证据路径。适用于实验方案、预实验复盘、结果台账或实验阻塞;不把方案生成视为实验执行。

FileKindSize
SKILL.en.mdmarkdown3.1 KB
SKILL.mdskill md2.6 KB
agents/openai.yamlconfig284 B

SKILL.en.md

View on GitHub
---
name: phd-experiment
description: Design testable research experiments, record actual runs and results, and maintain baselines and evidence paths. Use for experimental plans, pilot reviews, result logging, blockers or Update experiment progress. Generating a plan does not mean an experiment has run.
---

# Experimental design and results

Read the [local workflow](../phd-manager/references/local-workflow.en.md), then list projects, papers and experiments to obtain existing records and association IDs. Reply in English to English requests and preserve original research content unless a change or translation is requested.

## Design

Derive a falsifiable hypothesis, manipulated/compared factors, control or baseline, sample/data source, primary metric, evaluation conditions and failure criteria from the research question. Decide whether power analysis, repeated measurements, randomization, blinding, data splits or ablations apply to the discipline; do not impose a computational template on every study.

Put the executable plan in `hypothesis`, `method`, `metric` and `baseline`. Save longer plans as files and record their `artifactPath`. Mark unspecified budgets, sample sizes and expected effects as unresolved. A new AI-proposed plan uses `status: "planned"` and `origin: "ai"`; planned tasks can reference its `experimentId`.

## Runs and records

- A planned run stays `planned`. Use `running` only when the user explicitly reports a run or a tool has actually started it. Execution needs authorization, an available environment and an actual entry point. Invoking the skill does not start a background experiment.
- Before recording existing results, read supplied files, logs or explicit user statements. Preserve units, dataset/sample scope, baseline, settings, actual time and known anomalies. Do not claim to verify a missing or unreadable file.
- Put supported observations in `result` and their location and verification scope in `evidence`. Use `completed` only when actual results and evidence are both present. User-reported results can be labeled as not independently checked; never claim to have performed the run yourself.
- Failed startup, missing resources or a pending dependency can be `blocked`; describe the cause and next release condition in notes. Failures and negative results are valid records, not entries to remove for a better outcome.
- Charts, statistics and comparisons must follow the original data and actual calculations, with processing methods retained. Do not claim statistical significance without a test.

## Finish

Save through the CLI and read back. Report the plan/result location, what actually ran, the evidence supporting the conclusion and outstanding checks. Preserve existing entries when updating a paper's `experimentIds`. Completing an experiment does not automatically complete its paper or every associated task.

## Imported upstream resources

When the task benefits from the original methods or bundled tools, use the [GitHub integration](../github-experimental-design/SKILL.en.md) and read its complete upstream instructions. Check its requirements and scope; tool output does not replace the local record workflow.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall