Analyze Patterns

Analyze tool operation logs to identify repeated patterns across sessions and auto-generate reusable skills.

Pre-analysis data

!python3 $HOME/.claude/scripts/pre-analyze.py

Instructions

You have received a structured pre-analysis report above. Use it to evaluate patterns and make recommendations. Do NOT read the raw JSONL log file.

1. Review statistics

Briefly summarize the statistics: activity level, dominant tools, most active directories.

2. Scan existing skills

Read all .md files in ~/.claude/commands/ and all SKILL.md in ~/.claude/skills/*/ to understand what skills already exist.

3. Evaluate detected patterns

For each pattern, compare against existing skills:

• Already covered → Skip
• Partially overlapping → Suggest updating the existing skill
• New pattern → Suggest creating a new skill

For each actionable pattern, output:

### Pattern: <name>
- **Overlap**: [None / Partial with `<skill>` / Fully covered]
- **Recommendation**: [Create new / Update existing / Skip]
- **Frequency**: N sessions
- **Typical steps**: specific operations
- **Parameterizable**: what varies → use $ARGUMENTS
- **Suggested filename**: <name>.md
- **Suggested content**: (complete .md)

4. Create skills

Ask the user which to create/update. Write .md files to ~/.claude/commands/.

5. Log maintenance

If the summary mentions log cleanup needed, offer to trim (keep 30,000 lines).

Notes

• Generated skills use $ARGUMENTS for user-provided arguments
• Skill content should be clear instructions for Claude, not shell scripts
• Prioritize high-value patterns that save the most time
• If no patterns found, suggest using Claude Code for more sessions