Feedback Analyzer — Comprehensive Feedback Analysis Pipeline
Collects and analyzes customer/employee feedback data to produce sentiment analysis, topic classification, trend detection, and insight reports through agent team collaboration.
Execution Mode
Agent Team — 5 members communicate directly via SendMessage and cross-validate each other's work.
Agent Roster
| Agent |
File |
Role |
Type |
| data-collector |
.claude/agents/data-collector.md |
Data collection, cleansing, normalization |
general-purpose |
| sentiment-analyst |
.claude/agents/sentiment-analyst.md |
Sentiment analysis, emotion scoring |
general-purpose |
| topic-classifier |
.claude/agents/topic-classifier.md |
Topic classification, category design |
general-purpose |
| trend-detector |
.claude/agents/trend-detector.md |
Trend analysis, anomaly detection |
general-purpose |
| insight-writer |
.claude/agents/insight-writer.md |
Insight report, action items |
general-purpose |
Workflow
Phase 1: Preparation (performed directly by the orchestrator)
- Extract from user input:
- Feedback data: Text files, CSV, paste, or verbal description
- Data source type: Customer reviews/surveys/support logs/employee feedback/social media
- Analysis purpose: Product improvement/service quality/employee satisfaction/competitive analysis
- Comparison baseline (optional): Previous period, competitors, targets
- Existing categories (optional): User-defined classification system
- Create the
_workspace/ directory in the project root
- Organize the input and save to
_workspace/00_input.md
- Determine the execution mode based on the request scope
Phase 2: Team Assembly and Execution
| Order |
Task |
Owner |
Dependencies |
Deliverable |
| 1 |
Data collection/cleansing |
collector |
None |
_workspace/01_data_collection.md |
| 2a |
Sentiment analysis |
sentiment |
Task 1 |
_workspace/02_sentiment_analysis.md |
| 2b |
Topic classification |
classifier |
Task 1 |
_workspace/03_topic_classification.md |
| 3 |
Trend analysis |
trend |
Tasks 2a, 2b |
_workspace/04_trend_report.md |
| 4 |
Insight report |
writer |
Tasks 2a, 2b, 3 |
_workspace/05_insight_report.md |
Tasks 2a (sentiment) and 2b (topic) run in parallel. Both depend only on Task 1.
Inter-team communication flow:
- collector completes → sends cleansed data to sentiment, dataset/keywords to classifier, distribution data to trend
- sentiment completes → sends emotion-tagged data to classifier (for cross-analysis), time-series emotion data to trend
- classifier completes → sends topic time-series to trend, categories/urgent issues to writer
- trend completes → sends key trends/anomalies/forecasts to writer
- writer synthesizes all agent results into the insight report
Phase 3: Integration and Final Deliverables
- Verify all files in
_workspace/
- Cross-validation:
- Request corrections from the relevant agent if discrepancies are found (up to 2 rounds)
- Report the final summary to the user
Execution Modes by Scope
| User Request Pattern |
Execution Mode |
Agents Involved |
| "Analyze all feedback" |
Full Pipeline |
All 5 |
| "Just do sentiment analysis" |
Sentiment Mode |
collector + sentiment |
| "Classify this data by topic" |
Classification Mode |
collector + classifier |
| "Just look at trends" (existing analysis) |
Trend Mode |
trend + writer |
| "Just write the executive report" (existing analysis) |
Report Mode |
writer solo |
Leveraging existing analysis: When the user provides previous analysis results, copy to _workspace/ and skip the corresponding agent.
Data Transfer Protocol
| Strategy |
Method |
Usage |
| File-based |
_workspace/ directory |
Primary deliverable storage and sharing |
| Message-based |
SendMessage |
Real-time key information transfer, correction requests |
| Task-based |
TaskCreate/TaskUpdate |
Progress tracking, dependency management |
Error Handling
| Error Type |
Strategy |
| Data parsing failure |
collector processes text-extractable portions only, notes failure sections |
| Extremely small data (fewer than 5) |
Include "insufficient for statistical significance" warning, switch to qualitative analysis |
| Mixed languages |
Separate analysis by language, analyze primary language only if cross-comparison is not possible |
| High sentiment analysis uncertainty |
Tag with "[Confidence: Low]", recommend manual review |
| Agent failure |
1 retry → proceed without that deliverable if still failing, note omission in report |
Test Scenarios
Normal Flow
Prompt: "Analyze this customer review data. It's 200 app store reviews from the last 3 months."
Expected Results:
- Data collection: 200 entries cleansed, basic statistics, channel distribution
- Sentiment: Positive/negative/neutral ratios, NPS estimate, extreme sentiment highlights
- Topic: 3-5 major categories, topic × sentiment cross-analysis
- Trends: 3-month sentiment trajectory, anomaly detection
- Insights: Executive Summary, Top 3 insights, priority matrix
Existing File Flow
Prompt: "I have a previous sentiment analysis. Based on that, just create trends and a report."
Expected Results:
- Existing sentiment results copied to
_workspace/02_sentiment_analysis.md
- Trend mode: trend + writer only
- Trend analysis and insight report generated from existing data
Error Flow
Prompt: "Analyze just 5 customer feedback entries"
Expected Results:
- Include "insufficient for statistical significance" warning
- Switch to qualitative analysis: in-depth analysis of individual feedback
- Provide "snapshot analysis" instead of trend analysis
- Report specifies "Additional data collection recommended"
Agent Extension Skills
| Extension Skill |
Path |
Target Agent |
Role |
| sentiment-scoring |
.claude/skills/sentiment-scoring/skill.md |
sentiment-analyst |
Sentiment classification, scoring, NPS, context correction |
| text-analytics-methods |
.claude/skills/text-analytics-methods/skill.md |
topic-classifier, trend-detector |
Topic classification, keyword analysis, trend detection, insight derivation |
1---2name: feedback-analyzer3description: A comprehensive customer/employee feedback analysis pipeline. An agent team collaborates to handle data collection, sentiment analysis, topic classification, trend detection, and insight reporting. Use this skill for 'analyze feedback', 'customer review analysis', 'survey results analysis', 'VOC analysis', 'employee satisfaction analysis', 'NPS analysis', 'customer complaint analysis', 'feedback trends', 'sentiment analysis', and similar feedback/review/survey analysis topics. Survey design, customer response manual creation, and CRM system development are out of scope.4---56# Feedback Analyzer — Comprehensive Feedback Analysis Pipeline78Collects and analyzes customer/employee feedback data to produce sentiment analysis, topic classification, trend detection, and insight reports through agent team collaboration.910## Execution Mode1112**Agent Team** — 5 members communicate directly via SendMessage and cross-validate each other's work.1314## Agent Roster1516| Agent | File | Role | Type |17|-------|------|------|------|18| data-collector | `.claude/agents/data-collector.md` | Data collection, cleansing, normalization | general-purpose |19| sentiment-analyst | `.claude/agents/sentiment-analyst.md` | Sentiment analysis, emotion scoring | general-purpose |20| topic-classifier | `.claude/agents/topic-classifier.md` | Topic classification, category design | general-purpose |21| trend-detector | `.claude/agents/trend-detector.md` | Trend analysis, anomaly detection | general-purpose |22| insight-writer | `.claude/agents/insight-writer.md` | Insight report, action items | general-purpose |2324## Workflow2526### Phase 1: Preparation (performed directly by the orchestrator)27281. Extract from user input:29 - **Feedback data**: Text files, CSV, paste, or verbal description30 - **Data source type**: Customer reviews/surveys/support logs/employee feedback/social media31 - **Analysis purpose**: Product improvement/service quality/employee satisfaction/competitive analysis32 - **Comparison baseline** (optional): Previous period, competitors, targets33 - **Existing categories** (optional): User-defined classification system342. Create the `_workspace/` directory in the project root353. Organize the input and save to `_workspace/00_input.md`364. Determine the **execution mode** based on the request scope3738### Phase 2: Team Assembly and Execution3940| Order | Task | Owner | Dependencies | Deliverable |41|-------|------|-------|-------------|-------------|42| 1 | Data collection/cleansing | collector | None | `_workspace/01_data_collection.md` |43| 2a | Sentiment analysis | sentiment | Task 1 | `_workspace/02_sentiment_analysis.md` |44| 2b | Topic classification | classifier | Task 1 | `_workspace/03_topic_classification.md` |45| 3 | Trend analysis | trend | Tasks 2a, 2b | `_workspace/04_trend_report.md` |46| 4 | Insight report | writer | Tasks 2a, 2b, 3 | `_workspace/05_insight_report.md` |4748Tasks 2a (sentiment) and 2b (topic) run **in parallel**. Both depend only on Task 1.4950**Inter-team communication flow:**51- collector completes → sends cleansed data to sentiment, dataset/keywords to classifier, distribution data to trend52- sentiment completes → sends emotion-tagged data to classifier (for cross-analysis), time-series emotion data to trend53- classifier completes → sends topic time-series to trend, categories/urgent issues to writer54- trend completes → sends key trends/anomalies/forecasts to writer55- writer synthesizes all agent results into the insight report5657### Phase 3: Integration and Final Deliverables58591. Verify all files in `_workspace/`602. Cross-validation:61 - [ ] Sentiment analysis count matches data collection count62 - [ ] Topic classification unclassified rate is 10% or below63 - [ ] Insights cite data evidence64 - [ ] Action items follow SMART principles653. Request corrections from the relevant agent if discrepancies are found (up to 2 rounds)664. Report the final summary to the user6768## Execution Modes by Scope6970| User Request Pattern | Execution Mode | Agents Involved |71|---------------------|----------------|-----------------|72| "Analyze all feedback" | **Full Pipeline** | All 5 |73| "Just do sentiment analysis" | **Sentiment Mode** | collector + sentiment |74| "Classify this data by topic" | **Classification Mode** | collector + classifier |75| "Just look at trends" (existing analysis) | **Trend Mode** | trend + writer |76| "Just write the executive report" (existing analysis) | **Report Mode** | writer solo |7778**Leveraging existing analysis**: When the user provides previous analysis results, copy to `_workspace/` and skip the corresponding agent.7980## Data Transfer Protocol8182| Strategy | Method | Usage |83|----------|--------|-------|84| File-based | `_workspace/` directory | Primary deliverable storage and sharing |85| Message-based | SendMessage | Real-time key information transfer, correction requests |86| Task-based | TaskCreate/TaskUpdate | Progress tracking, dependency management |8788## Error Handling8990| Error Type | Strategy |91|-----------|----------|92| Data parsing failure | collector processes text-extractable portions only, notes failure sections |93| Extremely small data (fewer than 5) | Include "insufficient for statistical significance" warning, switch to qualitative analysis |94| Mixed languages | Separate analysis by language, analyze primary language only if cross-comparison is not possible |95| High sentiment analysis uncertainty | Tag with "[Confidence: Low]", recommend manual review |96| Agent failure | 1 retry → proceed without that deliverable if still failing, note omission in report |9798## Test Scenarios99100### Normal Flow101**Prompt**: "Analyze this customer review data. It's 200 app store reviews from the last 3 months."102**Expected Results**:103- Data collection: 200 entries cleansed, basic statistics, channel distribution104- Sentiment: Positive/negative/neutral ratios, NPS estimate, extreme sentiment highlights105- Topic: 3-5 major categories, topic × sentiment cross-analysis106- Trends: 3-month sentiment trajectory, anomaly detection107- Insights: Executive Summary, Top 3 insights, priority matrix108109### Existing File Flow110**Prompt**: "I have a previous sentiment analysis. Based on that, just create trends and a report."111**Expected Results**:112- Existing sentiment results copied to `_workspace/02_sentiment_analysis.md`113- Trend mode: trend + writer only114- Trend analysis and insight report generated from existing data115116### Error Flow117**Prompt**: "Analyze just 5 customer feedback entries"118**Expected Results**:119- Include "insufficient for statistical significance" warning120- Switch to qualitative analysis: in-depth analysis of individual feedback121- Provide "snapshot analysis" instead of trend analysis122- Report specifies "Additional data collection recommended"123124## Agent Extension Skills125126| Extension Skill | Path | Target Agent | Role |127|----------------|------|--------------|------|128| sentiment-scoring | `.claude/skills/sentiment-scoring/skill.md` | sentiment-analyst | Sentiment classification, scoring, NPS, context correction |129| text-analytics-methods | `.claude/skills/text-analytics-methods/skill.md` | topic-classifier, trend-detector | Topic classification, keyword analysis, trend detection, insight derivation |
revfactory