Answer questions about a generated analysis report - reads report artifacts, performs additional analysis if needed, and saves the answer with metadata
Answer questions about a generated analysis report - reads report artifacts, performs additional analysis if needed, and saves the answer with metadata
Read a generated analysis report (from /analyze-results) and answer a specific question about it. Can perform additional analysis on the underlying data if the existing report doesn't fully answer the question. Saves the answer as markdown with supporting files in a dedicated directory.
Usage
/answer-question What is the sample size?
/answer-question How does weight vary by position? ./analysis_1
If no analysis directory is specified, the skill will locate the most recent one automatically.
Workflow
1. Parse the Input
Extract two things from the user's input:
Question: The question to answer (required)
Analysis directory: An optional path to a specific analysis directory (e.g., ./analysis_3)
If no analysis directory is provided, check how many analysis directories contain a report.md:
import glob
import os
existing = sorted(glob.glob("./analysis_*"), key=os.path.getmtime, reverse=True)
with_reports = [d for d in existing if os.path.isfile(os.path.join(d, "report.md"))]
If zero directories have reports, tell the user to run /analyze-results first.
If exactly one directory has a report, use it automatically.
If multiple directories have reports, use AskUserQuestion to let the user pick. Read the first line of each report.md to build descriptive labels:
# Build option labels like "analysis_5 — Anchoring Bias Experiment (2026-02-06)"
options = []
for d in with_reports:
first_line = open(os.path.join(d, "report.md")).readline().strip().lstrip("# ")
options.append({"label": os.path.basename(d), "description": first_line})
Then ask:
Which analysis directory should I use?
with those options. Use the user's choice as analysis_dir.
2. Read the Report and Data
Load the key artifacts from the analysis directory:
report.md — The main analysis report. Read it fully.
results.csv — The raw data. Load it into pandas for potential follow-up analysis.
survey.md — The survey structure. Read it for context about question wording and types.
Also read any other files in the directory (mermaid diagrams, existing charts) to have full context.
3. Create Output Directory
Create a directory named with a slugified version of the question:
import re
import os
defslugify(text, max_length=60):
"""Convert text to a URL-friendly slug."""
text = text.lower().strip()
# Remove punctuation except hyphens
text = re.sub(r'[^\w\s-]', '', text)
# Replace whitespace with hyphens
text = re.sub(r'[\s_]+', '-', text)
# Remove leading/trailing hyphens
text = text.strip('-')
# Truncate to max_length at a word boundaryiflen(text) > max_length:
text = text[:max_length].rsplit('-', 1)[0]
return text
slug = slugify(question)
output_dir = f"{analysis_dir}/{slug}"
os.makedirs(output_dir, exist_ok=True)
For example:
"What is the sample size?" → analysis_1/what-is-the-sample-size/
"How does weight vary by position?" → analysis_1/how-does-weight-vary-by-position/
If the directory already exists, append a numeric suffix: what-is-the-sample-size-2.
4. Answer the Question
Attempt to answer the question in this order:
a) Check the existing report first. If the report already contains the answer, extract and summarize it. No additional analysis needed.
b) If the report doesn't fully answer the question, do additional analysis. Write and run Python scripts against results.csv to compute statistics, generate charts, or perform deeper analysis. Save any generated charts or data files to the output directory.
c) If the question requires information not available in the data, say so clearly and explain what data would be needed.
When performing additional analysis, save the Python script as analysis.py in the output directory for reproducibility.
5. Write the Answer
Write the answer as answer.md in the output directory:
# [Question text here]
[Clear, concise answer in 1-3 paragraphs]
## Supporting Evidence
[Tables, statistics, or references to charts that back up the answer]
## Additional Context
[Any caveats, limitations, or related observations — only if relevant]
Keep answers direct and focused. Lead with the answer, then provide evidence.
If charts or visualizations were generated, reference them with relative paths:
6. Write Metadata
Save a meta.json file in the output directory:
import json
from datetime import datetime, timezone
meta = {
"question": question,
"slug": slug,
"analysis_dir": analysis_dir,
"timestamp": datetime.now(timezone.utc).isoformat(),
"files": os.listdir(output_dir),
"required_additional_analysis": True# or False
}
withopen(f"{output_dir}/meta.json", "w") as f:
json.dump(meta, f, indent=2)
7. Generate HTML (optional)
If the answer includes charts or tables, convert to HTML for easy viewing:
Python script used for additional analysis (if any)
*.png
Charts or visualizations generated (if any)
Example
User runs:
/answer-question What is the average weight by position?
Skill finds ./analysis_1/ (most recent), reads report.md and results.csv, sees the report already has a position breakdown, extracts and reformats it, and produces:
