session-mine

name	session_mine
description	Analyze Claude Code session logs for patterns, inefficiencies, and improvement opportunities. Parses JSONL session files to find hot files, error patterns, and skill gaps.
metadata	{"version":"2.0","effort":"medium","auto-invocable":false,"category":"workflow","compatible-claude-code":{"when_to_use":"When analyzing session patterns or finding skill improvement opportunities","allowed-tools":["Agent","Bash"]}}

Session Mining Skill

Usage: /session_mine [days_back]

Analyze Claude Code session logs to find patterns, inefficiencies, and improvement opportunities. Runs in a forked agent context.

Process

1. Parse arguments from $ARGUMENTS:

   • days_back: number of days to analyze (default: 7)

2. Spawn an analysis agent using the Agent tool:

   • subagent_type: general-purpose
   • model: sonnet
   • Task: Run the mining script and interpret results.

   Agent prompt:

   Run: py ~/.claude/scripts/session_mine.py <days_back>
   
   Then analyze the output and produce a report (under 2500 chars):
   
   **Session Stats:**
   - Total sessions, avg duration, longest session
   
   **Tool Usage:**
   - Top 5 most-used tools with counts
   - Tools that are underused (e.g., Grep vs Bash grep)
   
   **Hot Files:**
   - Most-read files (candidates for CLAUDE.md context)
   - Most-edited files
   
   **Patterns:**
   - Recurring search terms → missing docs
   - Common error patterns → missing rules
   - Repeated workflows → skill candidates
   
   **Skill Gap Analysis:**
   Look for these patterns that indicate a skill needs updating:
   - User corrections after skill-generated output (e.g., "no, remove that shelf",
     "make the holes bigger") → skill has wrong defaults or missing options
   - Repeated manual fixes to skill output (e.g., always editing the same section
     after generation) → skill template needs updating
   - WebSearch/research done during skill use → skill is missing reference data
   - Same skill invoked 3+ times in one session with incremental fixes → skill
     workflow is incomplete or unclear
   - Tool errors during skill execution → skill has wrong commands or assumptions
   
   For each gap found, output:
   - Which skill file needs updating
   - What specific section/content to add or fix
   - Evidence (session count, correction count)
   
   **Recommendations:**
   - 3 actionable improvements for CLAUDE.md, skills, or hooks
   - Prioritize skill updates over new rules (skills compound, rules don't)
   

3. Return the report to the user.

Rules

• Always use a subagent — session logs can be large.
• If the mining script doesn't exist or fails, report the error clearly.
• Do not modify any files. Analysis only.