Analyze Customer Feedback

Customer Feedback Analysis Skill

Use this skill to analyze and summarize customer feedback from multiple sources over the last 7 days (or a custom time period). The skill fetches raw data from GitHub issues, NPS survey responses, email feedback, and subscription cancellation comments for you to analyze.

Important: For all tasks in this skill, use METABASE_API_KEY from the environment for NPS data (Metabase API) and GOOGLE_OAUTH_TOKEN for Gmail.

Instructions for the Agent

This skill fetches raw feedback data. You (the agent) are responsible for organizing it into a factual report. Your job is to identify themes, count their frequency, and cite specific sources. Do NOT editorialize, speculate on root causes, assign severity/priority, or make recommendations.

1. Filter out noise — Ignore feedback that is not substantive:

   • Auto-reply/out-of-office messages (e.g., "自动回复", "This is an automated response")
   • Spam or promotional content
   • Internal test messages
   • Bounce-back emails
   • Generic "thank you" acknowledgments without substance

2. Identify themes — Scan all four channels (GitHub issues, NPS comments, emails, churn cancellations) and group feedback items that describe the same topic into themes. A theme is any topic mentioned by 2+ users or across 2+ channels. Single-mention items should be listed under "Other" at the end.

3. Count frequency — For each theme, report:

   • Total number of mentions across all channels
   • Breakdown by channel (e.g., "3 GitHub issues, 2 NPS comments, 1 email")
   • Order themes by total mention count, highest first

4. Cite sources — For every item in a theme:

   • GitHub issues: include issue number as a link (e.g., [#123](https://github.com/YOUR_ORG/YOUR_REPO/issues/123)) and the issue title
   • NPS comments: quote the comment verbatim and include the score
   • Emails: include the sender name and a brief description of the content
   • Churn cancellations: quote the comment verbatim and include the plan type, e.g., Cancellation (build): "[verbatim quote]"

5. Note cross-channel convergence — When a theme appears in 2+ channels, state that as a fact (e.g., "This theme appeared in 3 GitHub issues and 2 NPS detractor comments"). Do not interpret what the convergence means.

6. Track week-over-week changes — Read the most recent previous report from reports/customer_feedback_summaries/ and compare:

   • Which themes are new (not in the previous report)
   • Which themes from the previous report recurred (with updated counts)
   • Which themes from the previous report did not recur
   • Report counts only. Do not use trend arrows, severity labels, or words like "improving," "worsening," or "resolved."

7. Format your report — Structure your output as described in the Output Format section below. Do not add sections beyond what is specified. Do not use emoji severity markers (🔴🟡🟢). Do not include sections titled "Recommended Actions," "Risks," "Open Questions," "Interpretation," or "Key Insight."

When to Use

Use this skill when you need to:

• Get a weekly summary of customer feedback
• Identify trending issues or complaints
• Track NPS scores and sentiment over time
• Understand why users are canceling subscriptions
• Prepare for product reviews or planning sessions
• Understand customer pain points across multiple channels

Data Sources

1. GitHub Issues (YOUR_ORG/YOUR_REPO) - Bug reports and feature requests from public users
2. NPS Survey Responses (Metabase API) - Net Promoter Score data with user comments
3. Email Feedback (Gmail) - Emails sent to your feedback address (set FEEDBACK_EMAIL)
4. Subscription Cancellations (Metabase API) - Cancellation comments from prod.your_subscriptions_table, grouped by plan type. Uses the same METABASE_API_KEY as NPS.

Prerequisites

1. GitHub CLI

The gh CLI must be authenticated with access to YOUR_ORG/YOUR_REPO.

# Check authentication status
gh auth status

# Login if needed
gh auth login

2. Metabase API (for NPS data)

The METABASE_API_KEY environment variable should be set in your ennvironment.

The script queries your-gcp-project-id.prod.your_nps_survey_responses_table via the Metabase API at https://your-metabase-instance.metabaseapp.com.

3. Gmail API

Gmail uses GOOGLE_OAUTH_TOKEN. The token needs gmail.readonly scope.

4. Install Dependencies

pip install -r requirements.txt

Note: The NPS fetcher uses only Python standard library (urllib.request), so no additional packages are needed for NPS data.

Usage

Basic Usage (analyze last 7 days)

cd skills/analyze_customer_feedback
python analyze_feedback.py

Custom Time Period

python analyze_feedback.py --days 14    # Last 14 days
python analyze_feedback.py --days 30    # Last 30 days

JSON Output (for programmatic use)

python analyze_feedback.py --json

Skip Sources (if not configured)

python analyze_feedback.py --skip-gmail     # Skip Gmail if not set up
python analyze_feedback.py --skip-nps       # Skip NPS if BigQuery not configured
python analyze_feedback.py --skip-github    # Skip GitHub issues
python analyze_feedback.py --skip-churn     # Skip churn cancellation comments

Run Individual Fetchers

# GitHub issues only
python fetch_github_issues.py --days 7 --json

# NPS data only
python fetch_nps_data.py --days 7 --json

# Gmail feedback only
python fetch_gmail_feedback.py --days 7 --json

# Churn cancellations only
python fetch_churn_data.py --days 7 --json

Script Output

The script outputs raw feedback data in a structured format:

1. Summary - Counts of feedback items from each source
2. GitHub Issues - Full issue data (title, body, labels, comments, URL)
3. NPS Comments - Score, OS, date, and verbatim comment for responses that included a comment
4. Email Feedback - Subject, sender, date, and full content
5. Churn Cancellation Comments - Plan type and verbatim cancellation comment, grouped by plan type

You (the agent) should organize this data into the report format below.

Report Output Format

Your report must follow this structure exactly. Do not add extra sections.

# Customer Feedback Report — [Date Range]

**Period:** [time period]
**Generated:** [date]

---

## Summary

- GitHub issues filed: [N]
- NPS responses with comments: [N] out of [N] total
- Emails received: [N] ([N] after filtering noise)
- Churn cancellations with comments: [N]

---

## Themes (ordered by total mention count)

### 1. [Theme Name] — [N] mentions
**Channels:** [N] GitHub issues, [N] NPS comments, [N] emails
**Previous report:** [Yes — N mentions / No]

- [#NNNN](link): [issue title] ([N] comments)
- [#NNNN](link): [issue title]
- NPS (score [N]): "[verbatim quote]"
- Email from [sender]: [brief description]
- Cancellation ([plan_type]): "[verbatim quote]"

### 2. [Theme Name] — [N] mentions
...

### Other (single-mention items)
- [#NNNN](link): [issue title]
- NPS (score [N]): "[verbatim quote]"
- ...

---

## NPS Comments

List all NPS responses that included a written comment, grouped by score band. Only include responses with non-empty comments. Include the score for context but do not calculate or report the NPS score itself.

### Scores 9-10
- (Score [N]): "[verbatim quote]"
- (Score [N]): "[verbatim quote]"

### Scores 7-8
- (Score [N]): "[verbatim quote]"

### Scores 0-6
- (Score [N]): "[verbatim quote]"
- (Score [N]): "[verbatim quote]"

---

## Churn Cancellation Comments

List all cancellation comments, grouped by plan type. Include only subscriptions canceled in the reporting period that have a non-empty comment.

### [plan_type] ([N] cancellations)
- "[verbatim quote]"
- "[verbatim quote]"

### [plan_type] ([N] cancellations)
- "[verbatim quote]"

---

## Email Summary

[N] emails received. [N] after filtering auto-replies, spam, and non-substantive messages.

1. **[sender name]** — [brief factual description of content]
2. **[sender name]** — [brief factual description of content]
...

---

## Week-over-Week

### New themes (not in previous report)
- [Theme Name] — [N] mentions
- [Theme Name] — [N] mentions

### Recurring themes (also in previous report)
- [Theme Name] — [N] mentions this week, [N] last week
- [Theme Name] — [N] mentions this week, [N] last week

### Themes from previous report that did not recur
- [Theme Name] — [N] mentions last week, 0 this week

Integration with Other Skills

After generating the report, you can:

Troubleshooting

Metabase API Errors

# Verify METABASE_API_KEY is set
echo $METABASE_API_KEY | head -c 10

# Test the API connection
curl -s -w "\nHTTP_CODE:%{http_code}" -H "x-api-key: $METABASE_API_KEY" "https://your-metabase-instance.metabaseapp.com/api/user/current"

GitHub CLI Errors

# Check token scopes
gh auth status

# Re-authenticate with required scopes
gh auth login --scopes read:org,repo

Example Script Output

The script outputs raw structured data like this (you then organize it into the Report Output Format above):

# Customer Feedback Data
**Period:** Last 7 days
**Generated:** 2026-02-02 17:30 UTC

## Summary
- GitHub Issues: 23
- NPS Responses: 89
- Email Feedback: 30
- Churn Cancellations: 144

## GitHub Issues
### #21234: Agent mode freezes on large files
- **URL:** https://github.com/YOUR_ORG/YOUR_REPO/issues/21234
- **State:** open
- **Created:** 2026-02-01T10:30:00Z
- **Comments:** 8
- **Labels:** bug, agent-mode
- **Body:**

When I open a file larger than 10MB, the agent freezes...

## NPS Survey Responses
### Score: 10
- **OS:** macOS
- **Date:** 2026-02-01
- **Comment:** Love the new agent mode! Makes coding so much faster.

### Score: 4
- **OS:** Windows
- **Date:** 2026-02-01
- **Comment:** SSH keeps disconnecting after a few minutes.
...

## Churn Cancellation Comments
144 cancellations with comments.

### build (72 cancellations)
- [example comment about usage frequency vs. cost]
- [example comment about switching to another tool]
...

### pro (48 cancellations)
- [example comment about pricing vs. feature usage]
...

Notes

• The script outputs progress to stderr and the data to stdout, allowing easy piping
• Full feedback comments are displayed without truncation for complete context
• The Gmail module filters for emails to/from the address in FEEDBACK_EMAIL