菜单
Design complete surveys from free text requirements - generates a multi-file study project with Survey, ScenarioList, AgentList, and a Makefile
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/expectedparrot/ep-skills --skill create-study命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Generate a qualitative interview guide as an EDSL Survey with QuestionInterview, based on Geiecke & Jaravel (2026) methodology
Publish a completed study to GitHub - creates a repo with the analysis report as README.md, all charts, data, and a reproducibility footer linking to Expected Parrot
Generate a comprehensive, self-contained analysis report for a conjoint (choice-based) study with executive summary, methodology, part-worth utilities, segment analysis, and limitations
基于 SOC 职业分类
正在显示 SKILL.md
| name | create-study |
| description | Design complete surveys from free text requirements - generates a multi-file study project with Survey, ScenarioList, AgentList, and a Makefile |
| allowed-tools | Read, Glob, Skill, AskUserQuestion, Write |
| arguments | research_question |
This skill generates a multi-file study project with EDSL objects (Survey, ScenarioList, AgentList, ModelList) based on a free text description of what the user wants the study to accomplish.
Example:
/create-study Do LLMs exhibit anchoring bias?
Every study is a directory containing these files:
<study_name>/
├── Makefile
├── study_survey.py
├── study_scenario_list.py
├── study_agent_list.py
├── study_model_list.py
└── create_results.py
| File | Purpose | Exports |
|---|---|---|
study_survey.py | Defines questions and survey with rules/memory | survey |
study_scenario_list.py | Defines scenario variations | scenario_list |
study_agent_list.py | Defines respondent personas | agent_list |
study_model_list.py | Defines which LLMs to use | model_list |
create_results.py | Imports all objects, runs survey, saves results | (script) |
Makefile | Build system for running the study | — |
Extract from the free text:
If the description is ambiguous or missing key details, use AskUserQuestion to clarify before generating code.
Use AskUserQuestion to ask where the user wants the study:
Question: "Where would you like the study files?"
Header: "Output"
Options:
1. "Write to directory (Recommended)" - "Create a study directory with all files based on the topic"
2. "Display only" - "Show the code in the conversation without saving"
If the user chooses to write to a directory:
anchoring_bias_study/, food_preferences_study/)Read the consolidated reference skill for detailed implementation guidance:
| Skill | When to Read |
|---|---|
| edsl-survey-reference | Question types, templating, rules, memory, helpers, visualization |
Use the Skill tool to invoke edsl-survey-reference, or Read the SKILL.md file by using Glob("**/edsl-survey-reference/SKILL.md") to locate it.
Based on requirements, determine:
Generate all files in the study directory. Each file is described below.
study_survey.pyDefines and exports a survey object. Contains all question definitions and survey construction.
from edsl import (
Survey,
QuestionMultipleChoice,
QuestionFreeText,
QuestionLinearScale,
)
# === QUESTIONS ===
q_cuisine = QuestionMultipleChoice(
question_name="favorite_cuisine",
question_text="Which cuisine do you enjoy most?",
question_options=["Italian", "Japanese", "Mexican", "Indian", "Thai"]
)
q_dish = QuestionFreeText(
question_name="favorite_dish",
question_text="What is your favorite {{ favorite_cuisine.answer }} dish?"
)
q_frequency = QuestionLinearScale(
question_name="frequency",
question_text="How often do you eat {{ favorite_cuisine.answer }} food?",
question_options=[1, 2, 3, 4, 5],
option_labels={1: "Rarely", 5: "Very Often"}
)
q_why = QuestionFreeText(
question_name="why_favorite",
question_text="Why do you particularly enjoy {{ favorite_cuisine.answer }} cuisine?"
)
# === SURVEY ===
survey = (Survey([q_cuisine, q_dish, q_frequency, q_why])
.set_full_memory_mode())
study_scenario_list.pyDefines and exports a scenario_list object. If no scenarios are needed, export an empty ScenarioList.
from edsl import Scenario, ScenarioList
scenario_list = ScenarioList([
Scenario({"topic": "artificial intelligence"}),
Scenario({"topic": "climate change"}),
Scenario({"topic": "remote work"}),
])
When no scenarios are needed:
from edsl import ScenarioList
scenario_list = ScenarioList()
study_agent_list.pyDefines and exports an agent_list object. If no agents are needed, export an empty AgentList.
from edsl import Agent, AgentList
agent_list = AgentList([
Agent(traits={"persona": "health-conscious millennial", "age": 28}),
Agent(traits={"persona": "traditional home cook", "age": 55}),
Agent(traits={"persona": "adventurous foodie", "age": 35}),
Agent(traits={"persona": "busy professional", "age": 42}),
])
When no agents are needed:
from edsl import AgentList
agent_list = AgentList()
study_model_list.pyDefines and exports a model_list object specifying which LLMs to use. If no specific models are needed, export a ModelList with a sensible default.
from edsl import ModelList, Model
model_list = ModelList([
Model("gpt-4o"),
Model("claude-3-5-sonnet-20241022"),
])
When only one model is needed (default):
from edsl import ModelList, Model
model_list = ModelList([Model("gpt-4o")])
create_results.pyImports from the other study files, builds the job, runs it, saves results as compressed JSON, then exports to CSV.
from study_survey import survey
from study_scenario_list import scenario_list
from study_agent_list import agent_list
from study_model_list import model_list
job = survey.by(scenario_list).by(agent_list).by(model_list)
results = job.run()
results.to_json("results.json.gz")
results.to_csv("results.csv")
Key rules for create_results.py:
.by() calls: survey.by(scenario_list).by(agent_list).by(model_list)results.json.gz (the primary artifact — compressed JSON with full fidelity)results.csv (derived convenience format)MakefileProvides a build target for running the study.
results.json.gz: create_results.py study_survey.py study_scenario_list.py study_agent_list.py study_model_list.py
python create_results.py
Key rules for the Makefile:
results.json.gzcreate_results.py and all four study component filescreate_results.py also produces results.csv as a side effect"I want to survey people about their food preferences across 5 cuisines with different persona types."
food_preferences_study/study_survey.py
from edsl import (
Survey,
QuestionMultipleChoice,
QuestionFreeText,
QuestionLinearScale,
)
q_cuisine = QuestionMultipleChoice(
question_name="favorite_cuisine",
question_text="Which cuisine do you enjoy most?",
question_options=["Italian", "Japanese", "Mexican", "Indian", "Thai"]
)
q_dish = QuestionFreeText(
question_name="favorite_dish",
question_text="What is your favorite {{ favorite_cuisine.answer }} dish?"
)
q_frequency = QuestionLinearScale(
question_name="frequency",
question_text="How often do you eat {{ favorite_cuisine.answer }} food?",
question_options=[1, 2, 3, 4, 5],
option_labels={1: "Rarely", 5: "Very Often"}
)
q_why = QuestionFreeText(
question_name="why_favorite",
question_text="Why do you particularly enjoy {{ favorite_cuisine.answer }} cuisine?"
)
survey = (Survey([q_cuisine, q_dish, q_frequency, q_why])
.set_full_memory_mode())
food_preferences_study/study_scenario_list.py
from edsl import ScenarioList
scenario_list = ScenarioList()
food_preferences_study/study_agent_list.py
from edsl import Agent, AgentList
agent_list = AgentList([
Agent(traits={"persona": "health-conscious millennial", "age": 28}),
Agent(traits={"persona": "traditional home cook", "age": 55}),
Agent(traits={"persona": "adventurous foodie", "age": 35}),
Agent(traits={"persona": "busy professional", "age": 42}),
])
food_preferences_study/study_model_list.py
from edsl import ModelList, Model
model_list = ModelList([Model("gpt-4o")])
food_preferences_study/create_results.py
from study_survey import survey
from study_scenario_list import scenario_list
from study_agent_list import agent_list
from study_model_list import model_list
job = survey.by(scenario_list).by(agent_list).by(model_list)
results = job.run()
results.to_json("results.json.gz")
results.to_csv("results.csv")
food_preferences_study/Makefile
results.json.gz: create_results.py study_survey.py study_scenario_list.py study_agent_list.py study_model_list.py
python create_results.py
All three component files are still created. study_scenario_list.py and study_agent_list.py export empty lists.
study_survey.py uses {{ scenario.variable }} templating. study_scenario_list.py defines the variations.
study_survey.py may reference {{ agent.trait }}. study_agent_list.py defines the personas.
Both study_scenario_list.py and study_agent_list.py define their respective objects. create_results.py chains all: survey.by(scenario_list).by(agent_list).by(model_list).
study_model_list.py defines multiple models to compare responses across LLMs. Results are crossed with scenarios and agents.
Generate a study directory containing:
study_survey.py — imports EDSL question/survey classes, defines questions, creates and exports surveystudy_scenario_list.py — imports EDSL scenario classes, creates and exports scenario_list (empty if not needed)study_agent_list.py — imports EDSL agent classes, creates and exports agent_list (empty if not needed)study_model_list.py — imports EDSL model classes, creates and exports model_listcreate_results.py — imports from the four study files, runs the survey, saves results.json.gz and results.csvMakefile — build target for results.json.gz with all dependenciessurvey, scenario_list, agent_list, model_listcreate_results.py imports from all four study filescreate_results.py saves both results.json.gz and results.csv.py files as dependencies