Session Insights
Analyze the current conversation session and produce an improvement report.
Taxonomy and classification reference: ${CLAUDE_SKILL_DIR}/references/taxonomy.md
1. Session Overview
Assess the session and output:
## Session Overview
| Metric | Value |
| ------------ | ------------------------------------- |
| Category | <one from taxonomy reference> |
| Session Size | <XS/S/M/L/XL from taxonomy reference> |
| Turns | <integer> |
| Tools Used | <comma-separated list> |
| Skills Used | <comma-separated list, or "None"> |
2. Issue Timeline
Scan the conversation for issue types from taxonomy reference.
Output as timeline:
## Issue Timeline
1. **[Turn <N>]** <Issue Type from taxonomy reference> (Impact: high/medium/low)
<What happened> → <What the correct action was>
If no issues found: "No issues identified. [one-sentence reasoning why session was efficient]"
3. Actionable Feedback
At least one of the following subsections must be present. Omit subsections only when genuinely not applicable.
Prompt Improvement
If the initial prompt could have prevented issues:
### Improved Prompt
> [Rewritten version of the user's initial request]
**Changes:**
- [What was added/clarified and why]
CLAUDE.md / Rules Suggestions
If the session revealed missing conventions, preferences, or patterns:
### Suggested Additions
**CLAUDE.md:**
- [Rule or preference to add]
**.claude/rules/[topic].md:**
- [Rule to add, with paths pattern]
Only suggest additions that would prevent issues observed in THIS session. Do not suggest generic improvements.
Skill Suggestions
If a skill was missing, underperformed, or misused:
### Skill Feedback
- **[skill-name]**: [What went wrong and how to improve the skill]
- **Missing skill**: [Description of a skill that would have helped]
4. Knowledge Usage
Review which CLAUDE.md rules and .claude/rules/ were relevant:
## Knowledge Usage
### Applied
- [Rule/instruction that was followed and helped]
### Missed
- [Rule/instruction that existed but was not applied when it should have been]
### Misleading
- [Rule/instruction that caused incorrect behavior — consider updating]
Read CLAUDE.md and relevant .claude/rules/ files to cross-reference.
Guidelines
- Most valuable for longer sessions (16+ turns) where patterns emerge
- Be specific — cite conversation turns, not vague observations
- Focus on actionable improvements, not praise
- L/XL sessions deserve deeper analysis
- For XS/S sessions with no issues, keep the report brief