Collect Feedback

Gather structured feedback from the user about their experience with aiframework-generated output. This data feeds into /aif-evolve for continuous improvement.

When to Use

• After a user has run aiframework on a project
• When a user reports dissatisfaction with generated output
• Periodically to collect improvement signals

Step 0: Telemetry (silent, non-blocking)

aiframework-telemetry skill_invoked skill=aif-feedback 2>/dev/null || true

Step 1: Ask 5 Questions

Present these questions one at a time, waiting for each answer:

1. CLAUDE.md Quality (1-5): "How useful is the generated CLAUDE.md for your daily work?"
2. Accuracy (1-5): "How accurate were the detected stack, domains, and commands?"
3. Missing Context (free text): "What important context about your project is missing from the output?"
4. Noise (free text): "What generated content is unhelpful or distracting that you'd remove?"
5. Top Wish (free text): "If you could improve one thing about aiframework, what would it be?"

Step 2: Save Feedback

Append a JSON line to tools/learnings/feedback.jsonl:

jq -n --arg d "$(date +%Y-%m-%d)" --arg q SCORE --arg a SCORE --arg mc "ANSWER" --arg n "ANSWER" --arg tw "ANSWER" --arg p "$(basename "$(pwd)")" \
  '{date:$d, claude_md_quality:($q|tonumber), accuracy:($a|tonumber), missing_context:$mc, noise:$n, top_wish:$tw, project:$p}' >> tools/learnings/feedback.jsonl

Step 3: Acknowledge

Thank the user and explain:

• Feedback is stored locally in tools/learnings/feedback.jsonl
• Running /aif-evolve will incorporate this feedback into improvement recommendations
• Patterns across feedback entries drive CLAUDE.md and rule evolution

Integration with /aif-evolve

The evolve skill reads feedback.jsonl alongside *-learnings.jsonl to:

• Identify recurring complaints (noise, missing context)
• Prioritize improvements by frequency and severity
• Track quality scores over time