菜单
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
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 publish-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。
正在显示 SKILL.md
基于 SOC 职业分类
| name | publish-study |
| description | 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 |
| allowed-tools | Read, Glob, Grep, Bash(python3:*), Bash(gh *), Bash(git *), Bash(cp *), Bash(mkdir *), Write, AskUserQuestion |
| user-invocable | true |
| arguments | study_dir |
Takes a completed study directory (containing results, an experiment design, and an analysis report) and publishes it as a GitHub repository. The analysis report becomes README.md so it renders nicely on GitHub.
/publish-study
/publish-study 2026-02-08_do-llms-exhibit-maternal-default-bias-when
If no study directory is specified, auto-detect from the current working directory by looking for directories matching YYYY-MM-DD_*.
If a directory was provided, confirm it exists. Otherwise, find candidates:
import glob
import os
candidates = sorted(glob.glob(), key=os.path.getmtime, reverse=)
candidates = [d d candidates os.path.isdir(d)]
AskUserQuestion to let the user pick:Question: "Which study do you want to publish?"
Header: "Study"
Options: [list each directory as an option with its name as label and first line of experiment_design.md as description]
Find analysis_N/ subdirectories that contain report.md:
import os, glob
analysis_dirs = sorted(glob.glob(f"{study_dir}/analysis_*/"))
with_reports = [d for d in analysis_dirs if os.path.isfile(os.path.join(d, "report.md"))]
/analyze-results first.AskUserQuestion to let the user pick which analysis to feature as the README:Question: "Which analysis should be the primary report (README.md)?"
Header: "Analysis"
Options: [list each analysis_N with first line of its report.md as description]
Check for results.json.gz in the study directory and push it to Expected Parrot so the repo can link to it for reproducibility.
from edsl import Results
results = Results.load(f"{study_dir}/results") # loads results.json.gz
# Push to Expected Parrot (unlisted by default)
info = results.push(description="Results for: <study_slug>")
# Extract the UUID and URL from the push response
# info is a dict with 'uuid' and 'url' keys
results_uuid = info["uuid"]
results_url = info["url"]
print(f"Results pushed: {results_url}")
print(f"UUID: {results_uuid}")
If the push fails or results.json.gz doesn't exist, warn the user but continue — the reproducibility section will note that results are not available on Expected Parrot.
Build the repository contents in a temporary directory. Copy files from the study and analysis directories:
# Create temp staging directory
mkdir -p /tmp/publish-study-staging
mkdir -p /tmp/publish-study-staging/images
mkdir -p /tmp/publish-study-staging/data
mkdir -p /tmp/publish-study-staging/questions
mkdir -p /tmp/publish-study-staging/study
| Source | Destination in Repo | Notes |
|---|---|---|
analysis_N/report.md | README.md | Rewrite image/data paths, append reproducibility footer |
analysis_N/*.png | images/*.png | All chart PNGs from the analysis |
study_*.py | study/study_*.py | All EDSL component files (survey, agents, models, scenarios) |
create_results.py | study/create_results.py | Results runner script |
analyze_*.py | study/analyze_*.py | Analysis scripts |
Makefile | study/Makefile | Build system |
design_spec.json | study/design_spec.json | Experimental design spec |
conjoint_choice_sets.json | study/conjoint_choice_sets.json | Choice sets (if conjoint) |
*.md (top-level, not README) | study/*.md | Design docs (conjoint_design.md, experiment_design.md) |
analysis_N/survey.md | survey.md | Survey documentation |
analysis_N/results.csv | data/results.csv | Tabular results |
analysis_N/<slug>/answer.md | questions/<slug>.md | Answer-question outputs (if any) |
analysis_N/<slug>/*.png | questions/<slug>/*.png | Answer charts (if any) |
| — | LICENSE | MIT license (generated) |
STUDY_DIR="<study_dir>"
ANALYSIS_DIR="<analysis_dir>"
STAGING="/tmp/publish-study-staging"
# Clean staging area
rm -rf "$STAGING"
mkdir -p "$STAGING/images" "$STAGING/data" "$STAGING/questions" "$STAGING/study"
# Copy analysis PNGs to images/
cp "$ANALYSIS_DIR"/*.png "$STAGING/images/" 2>/dev/null || true
# Copy all study Python files into study/ directory
cp "$STUDY_DIR"/study_*.py "$STAGING/study/" 2>/dev/null || true
cp "$STUDY_DIR"/create_results.py "$STAGING/study/" 2>/dev/null || true
cp "$STUDY_DIR"/analyze_*.py "$STAGING/study/" 2>/dev/null || true
# Copy Makefile
cp "$STUDY_DIR"/Makefile "$STAGING/study/" 2>/dev/null || true
# Copy JSON design files
cp "$STUDY_DIR"/design_spec.json "$STAGING/study/" 2>/dev/null || true
cp "$STUDY_DIR"/conjoint_choice_sets.json "$STAGING/study/" 2>/dev/null || true
# Copy design/experiment docs (top-level .md files, excluding README)
for md in "$STUDY_DIR"/*.md; do
[ -f "$md" ] && cp "$md" "$STAGING/study/" 2>/dev/null || true
done
# Copy survey documentation
cp "$ANALYSIS_DIR/survey.md" "$STAGING/survey.md" 2>/dev/null || true
# Copy results CSV
cp "$ANALYSIS_DIR/results.csv" "$STAGING/data/results.csv" 2>/dev/null || true
Check for subdirectories inside the analysis directory that contain answer.md:
import glob, os, shutil
staging = "/tmp/publish-study-staging"
analysis_dir = "<analysis_dir>"
answer_dirs = [d for d in glob.glob(f"{analysis_dir}/*/")
if os.path.isfile(os.path.join(d, "answer.md"))]
for answer_dir in answer_dirs:
slug = os.path.basename(answer_dir.rstrip("/"))
dest = f"{staging}/questions/{slug}"
os.makedirs(dest, exist_ok=True)
# Copy answer.md as <slug>.md
shutil.copy2(f"{answer_dir}/answer.md", f"{staging}/questions/{slug}.md")
# Copy any PNGs
for png in glob.glob(f"{answer_dir}/*.png"):
os.makedirs(f"{dest}", exist_ok=True)
shutil.copy2(png, f"{dest}/{os.path.basename(png)}")
Read the original report.md and apply three modifications:
import re
with open(f"{analysis_dir}/report.md", "r") as f:
readme = f.read()
# 1. Rewrite image paths: (foo.png) -> (images/foo.png)
# Match markdown image syntax 
readme = re.sub(
r'(!\[[^\]]*\])\(([^/)][^)]*\.png)\)',
r'\1(images/\2)',
readme
)
# 2. Rewrite results.csv links: (results.csv) -> (data/results.csv)
readme = readme.replace("(results.csv)", "(data/results.csv)")
# 3. Append reproducibility footer
footer = f"""
---
## Reproducibility
This study was conducted using [Expected Parrot EDSL](https://docs.expectedparrot.com/).
### Results on Expected Parrot
The full results object is available on Expected Parrot:
- **URL:** [{results_url}]({results_url})
- **UUID:** `{results_uuid}`
### Pull the Results
```python
from edsl import Results
results = Results.pull("{results_uuid}")
df = results.to_pandas()
The full study code is available in the study/ directory, including:
This study and its artifacts are released under the MIT License.
Generated with Expected Parrot EDSL """
readme += footer
with open(f"{staging}/README.md", "w") as f: f.write(readme)
If the EP push failed (no `results_uuid`), use a simpler footer without the EP link:
```python
footer_no_ep = """
---
## Reproducibility
This study was conducted using [Expected Parrot EDSL](https://docs.expectedparrot.com/).
### Study Code
The full study code is available in the [`study/`](study/) directory, including:
- EDSL component definitions (survey, agents, models, scenarios)
- Results runner script
- Makefile for reproducibility
- Experimental design specification
### License
This study and its artifacts are released under the [MIT License](LICENSE).
---
*Generated with [Expected Parrot EDSL](https://docs.expectedparrot.com/)*
"""
Write a standard MIT license file:
from datetime import datetime
year = datetime.now().year
license_text = f"""MIT License
Copyright (c) {year}
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
"""
with open(f"{staging}/LICENSE", "w") as f:
f.write(license_text)
Use AskUserQuestion to confirm the repo name, visibility, and GitHub account:
Question: "What should the GitHub repo be named?"
Header: "Repo name"
Options:
1. "<study-slug>" - "Default: slug from the study directory name (without date prefix)"
2. "Custom name" - "Enter a custom repository name"
Then ask about visibility:
Question: "Should the repo be public or private?"
Header: "Visibility"
Options:
1. "Public (Recommended)" - "Anyone can see this repository"
2. "Private" - "Only you and collaborators can see this repository"
The study slug is derived from the directory name by stripping the date prefix:
# "2026-02-08_do-llms-exhibit-maternal-default-bias-when" -> "do-llms-exhibit-maternal-default-bias-when"
study_slug = os.path.basename(study_dir).split("_", 1)[1] if "_" in os.path.basename(study_dir) else os.path.basename(study_dir)
REPO_NAME="<repo_name>"
VISIBILITY="--public" # or "--private"
STAGING="/tmp/publish-study-staging"
# Create the GitHub repo
gh repo create "$REPO_NAME" $VISIBILITY --description "Research study: <short description>"
# Initialize git in the staging directory and push
cd "$STAGING"
git init
git add .
git commit -m "Initial commit: publish study results
Generated with Expected Parrot EDSL
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>"
# Add remote and push
git remote add origin "$(gh repo view "$REPO_NAME" --json url -q .url)"
git branch -M main
git push -u origin main
Tell the user:
Question: (none — just report)
Example output:
Published to: https://github.com/<user>/demand-and-price-sensitivity-for-frozen-chicken
Repository contents:
- README.md (analysis report with 3 charts)
- survey.md
- study/ (11 files: Python scripts, Makefile, design specs, docs)
- data/results.csv
- images/ (3 PNGs)
- LICENSE (MIT)
Results are also available on Expected Parrot:
https://expectedparrot.com/content/<uuid>
Given a study directory:
2026-02-11_demand-and-price-sensitivity-for-frozen-chicken/
├── experiment_design.md
├── conjoint_design.md
├── design_spec.json
├── conjoint_choice_sets.json
├── study_survey.py
├── study_agent_list.py
├── study_model_list.py
├── study_scenario_list.py
├── create_results.py
├── analyze_conjoint.py
├── Makefile
├── results.json.gz
├── results.csv
└── analysis_1/
├── report.md
├── report.html
├── survey.md
├── results.csv
├── part_worth_utilities.png
├── attribute_importance.png
└── segment_analysis.png
The published repo would contain:
demand-and-price-sensitivity-for-frozen-chicken/
├── README.md # report.md with rewritten paths + reproducibility footer
├── survey.md
├── LICENSE
├── data/
│ └── results.csv
├── images/
│ ├── part_worth_utilities.png
│ ├── attribute_importance.png
│ └── segment_analysis.png
└── study/
├── study_survey.py
├── study_agent_list.py
├── study_model_list.py
├── study_scenario_list.py
├── create_results.py
├── analyze_conjoint.py
├── Makefile
├── design_spec.json
├── conjoint_choice_sets.json
├── experiment_design.md
└── conjoint_design.md
results.json.gzanalysis_N/ directory has a report.mdstudy/ directory contains Python files (study_*.py, create_results.py, etc.)study/ directory contains design docs (experiment_design.md, design_spec.json, etc.)images/data/results.csv| File | Description |
|---|---|
README.md | Analysis report adapted for GitHub with image paths rewritten and reproducibility footer |
study/study_*.py | EDSL component definitions (survey, agents, models, scenarios) |
study/create_results.py | Results runner script |
study/analyze_*.py | Analysis scripts |
study/Makefile | Build system for reproducibility |
study/design_spec.json | Experimental design specification |
study/conjoint_choice_sets.json | Choice set definitions (if conjoint) |
study/*.md | Design docs (experiment_design.md, conjoint_design.md) |
survey.md | Survey documentation (questions, types, options) |
data/results.csv | Tabular results data |
images/*.png | All analysis chart PNGs |
questions/<slug>.md | Answer-question outputs (if any) |
questions/<slug>/*.png | Answer-question charts (if any) |
LICENSE | MIT license |
/analyze-results before /publish-study — you need at least one analysis with a report.mdanalysis_N/report.md firstgh CLI is not authenticated, run gh auth login first