Feedback Skill

Overview

Simple feedback system for content quality. Collect ratings, identify patterns, and get recommendations for rule updates.

Philosophy: Lightweight pattern identification → Manual rule improvements

Operations

rate - Collect feedback on content quality

• Entry: feedback-skill rate <path>
• Collects: Rating (1-5), issue category, optional comment
• Output: Stores in .kurt/kurt.sqlite
• Subskill: subskills/rate.md

dashboard - View feedback trends and summary

• Entry: feedback-skill dashboard [--days <n>]
• Shows: Overall stats, issue breakdown, rating trends, recent feedback
• Output: Console summary
• Subskill: subskills/dashboard.md

patterns - Identify recurring issues and recommend updates

• Entry: feedback-skill patterns [--days <n>] [--min-frequency <n>]
• Shows: Issues that occur ≥3 times with recommendations
• Output: Recommended writing-rules-skill commands
• Subskill: subskills/patterns.md

Routing Logic

Parse arguments → Route to subskill:

OPERATION=$1
shift

case "$OPERATION" in
    "rate")
        .claude/skills/feedback-skill/subskills/rate.md "$@"
        ;;

    "dashboard")
        .claude/skills/feedback-skill/subskills/dashboard.md "$@"
        ;;

    "patterns")
        .claude/skills/feedback-skill/subskills/patterns.md "$@"
        ;;

    *)
        echo "Unknown operation: $OPERATION"
        echo ""
        echo "Available operations:"
        echo "  rate       - Rate content quality"
        echo "  dashboard  - View feedback trends"
        echo "  patterns   - Identify recurring issues"
        exit 1
        ;;
esac

Data Storage

SQLite (.kurt/kurt.sqlite)

feedback_events (simplified schema)

CREATE TABLE feedback_events (
    id TEXT PRIMARY KEY,           -- UUID
    created_at TEXT NOT NULL,      -- ISO 8601 timestamp
    rating INTEGER NOT NULL,       -- 1-5
    comment TEXT,                  -- Optional text feedback
    issue_category TEXT,           -- tone|structure|info|comprehension|length|examples|other
    asset_path TEXT,               -- Path to rated content
    project_id TEXT                -- Optional project context
);

Removed tables (from previous complex version):

• improvements - No automated execution tracking
• workflow_retrospectives - Workflows removed
• workflow_phase_ratings - Workflows removed
• feedback_loops - Too complex for simple system

Simple Feedback Flow

1. User creates content

2. User rates content (optional):
   feedback-skill rate <path>
   → Rating: 1-5
   → Issue category (if ≤3)
   → Optional comment

3. View trends over time:
   feedback-skill dashboard
   → Overall stats
   → Issue breakdown
   → Rating trends

4. When patterns emerge (≥3 occurrences):
   feedback-skill patterns
   → Shows recurring issues
   → Recommends rule update commands

5. User manually updates rules:
   writing-rules-skill style --type X --update
   writing-rules-skill structure --type X --update
   writing-rules-skill persona --audience-type X --update

No automation. User decides when to act.

Issue Categories

Simple, content-focused categories:

Integration Points

From content-writing-skill

Optional integration (not required):

# After draft creation
echo ""
echo "Rate this draft? (y/N): "
read -r RESPONSE

if [ "$RESPONSE" = "y" ] || [ "$RESPONSE" = "Y" ]; then
    feedback-skill rate "$DRAFT_PATH"
fi

To writing-rules-skill

patterns.md recommends commands:

Tone Issues (5× in last 30 days)
→ writing-rules-skill style --type technical-docs --update

Structure Issues (3× in last 30 days)
→ writing-rules-skill structure --type tutorial --update

User copies and runs command to update rules.

Configuration

Minimal configuration in .kurt/feedback/feedback-config.yaml:

feedback:
  enabled: true
  min_pattern_frequency: 3  # Minimum occurrences to show pattern
  default_time_window_days: 30

Removed from config:

• Issue mappings with automated commands (too complex)
• Improvement execution settings (no automation)
• Workflow-related configuration (workflows removed)

Design Principles

1. Simple and lightweight: Just collect → analyze → recommend
2. Pattern-based: Only show issues that occur multiple times
3. Manual execution: User runs update commands (no automation)
4. Non-blocking: Feedback collection never interrupts workflow
5. Content-focused: Only content quality (no projects/workflows)
6. Privacy-conscious: Minimal data storage

Example Usage

Rate a draft:

feedback-skill rate projects/my-tutorial/draft.md

View feedback trends:

feedback-skill dashboard
feedback-skill dashboard --days 7

Check for patterns:

feedback-skill patterns
feedback-skill patterns --min-frequency 5

After pattern identified, update rules:

# Copy recommended command from patterns output
writing-rules-skill style --type technical-docs --update

Getting Started

1. Create content (drafts, outlines)

2. Rate content occasionally:

   bashCopy

   feedback-skill rate path/to/draft.md
   

3. After several ratings, check dashboard:

   bashCopy

   feedback-skill dashboard
   

4. When patterns emerge, check recommendations:

   bashCopy

   feedback-skill patterns
   

5. Update rules based on patterns:

   bashCopy

   writing-rules-skill style --type X --update
   

What Changed from Previous Version

Removed:

• Project plan feedback (Loop 2)
• Workflow retrospectives (Loop 3)
• Automated improvement execution
• Validation and effectiveness tracking
• Complex feedback loop completion metrics
• Multiple feedback types (now just content quality)

Kept (simplified):

• Content rating with issue identification
• Pattern analysis across feedback
• Trend visualization in dashboard
• Manual rule update recommendations

Result: ~2,000 fewer lines of code, simpler user experience

This skill provides lightweight feedback collection and pattern analysis to guide manual rule improvements.