菜单
Analyze session logs to discover automation opportunities. Use periodically to find patterns in your Claude Code usage that could be automated. Use periodically after multiple sessions to find automation opportunities.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/majiayu000/claude-skill-registry --skill analyze-sessions命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | analyze-sessions |
| description | Analyze session logs to discover automation opportunities. Use periodically to find patterns in your Claude Code usage that could be automated. Use periodically after multiple sessions to find automation opportunities. |
| argument-hint | [--project <name>] [--targets-only] |
| allowed-tools | Read, Write, Glob, Grep, Bash |
Analyze recorded session data from ALL Claude Code projects to identify patterns and automation opportunities.
--project <name> — Analyze only sessions from the named project (matches the project field, which is the basename of the working directory)--targets-only — Exclude toolkit sessions; only analyze target projectsSession logs are stored centrally in ~/.claude/logs/sessions.jsonl in JSONL format (one compact JSON object per line).
Each entry contains:
{"session_id":"...","timestamp":"2026-01-23T12:00:00Z","project":
The session-end-logger hook is installed at the user level (~/.claude/hooks/session-end-logger.sh) so it captures sessions from every project.
Copy this checklist and track progress:
Analyze Sessions Progress:
- [ ] Step 1: Read session logs
- [ ] Step 2: Calculate statistics (grouped by project)
- [ ] Step 3: Identify target projects
- [ ] Step 4: Analyze transcripts
- [ ] Step 5: Find automation opportunities
- [ ] Step 6: Cross-project insights
- [ ] Step 7: Generate report
Read ~/.claude/logs/sessions.jsonl and parse all entries.
If --project <name> was given, filter to entries where project matches.
If --targets-only was given, identify toolkit projects (see step 3) and exclude them.
If the file doesn't exist or is empty:
NO SESSION DATA
===============
No sessions have been logged yet.
Session logging is installed at the user level (~/.claude/hooks/session-end-logger.sh).
Run a few Claude Code sessions in any project, then run /analyze-sessions again.
If the hook is not installed, run /install-hooks and select "session-end-logger".
SESSION ANALYSIS
================
Total sessions: {N}
Date range: {earliest} to {latest}
Unique projects: {count}
Sessions per project:
- {project1}: {count} sessions ({first_date} – {last_date})
- {project2}: {count} sessions ({first_date} – {last_date})
Scan ~/Projects/ for directories containing .claude/toolkit-version.json.
These are toolkit target projects. All other projects are either the toolkit itself or standalone.
Project Classification:
Toolkit: {toolkit project name}
Target projects: {list of target project names}
Other: {list of other project names}
For each session with an accessible transcript_path:
Manual Interventions:
AskUserQuestion tool calls — Questions asked during sessionsCommon Blockers:
Tool Usage Patterns:
Workflow Friction:
Based on patterns found, identify:
High Impact (Repeated across 3+ sessions):
Medium Impact (Repeated 2 times):
Low Impact (Single occurrence but notable):
Compare patterns across projects:
CROSS-PROJECT COMPARISON
========================
| Project | Sessions | Avg Questions/Session | Top Blocker | Autonomy Level |
|---------|----------|-----------------------|-------------|----------------|
| {proj} | {N} | {avg} | {blocker} | {High/Med/Low} |
Autonomy Level = High (0-1 questions/session), Med (2-3), Low (4+)
Identify:
Before writing the report file, output a summary of key findings to the user:
SESSION ANALYSIS PREVIEW
========================
Sessions analyzed: {N}
Top patterns found: {count}
Key findings:
1. {finding 1}
2. {finding 2}
3. {finding 3}
Writing full report to .claude/logs/ANALYSIS_REPORT.md...
This gives the user visibility before the file is written.
Write analysis to .claude/logs/ANALYSIS_REPORT.md (in the toolkit):
# Session Analysis Report
Generated: {timestamp}
Sessions analyzed: {N}
Date range: {earliest} to {latest}
## Summary
- Total sessions: {N}
- Projects: {list}
- Common patterns identified: {count}
## Per-Project Breakdown
### {Project Name} ({N} sessions)
**Type:** Toolkit | Target | Other
**Date range:** {first} – {last}
**Key patterns:**
- {pattern 1}
- {pattern 2}
## Cross-Project Insights
### Autonomy Levels
| Project | Sessions | Questions/Session | Autonomy |
|---------|----------|-------------------|----------|
| {proj} | {N} | {avg} | {level} |
### Shared Patterns
- {pattern that appears across multiple projects}
## Automation Opportunities
### High Priority
#### 1. {Pattern Name}
**Occurrences:** {N} sessions
**Projects:** {list of projects where this appears}
**Pattern:** {description of what keeps happening}
**Suggested Automation:**
- {specific automation approach}
- {implementation hint}
#### 2. {Pattern Name}
...
### Medium Priority
#### 1. {Pattern Name}
...
### Low Priority
#### 1. {Pattern Name}
...
## Recommended Actions
1. **Create new skill:** {skill name} — {what it would automate}
2. **Add to AGENTS.md:** {guidance to add}
3. **Create hook:** {hook description}
## Raw Statistics
### Questions Asked (AskUserQuestion)
| Question Pattern | Count | Projects |
|-----------------|-------|----------|
| {pattern} | {N} | {list} |
### Tools Used
| Tool | Total Uses | Avg per Session |
|------|------------|-----------------|
| {tool} | {N} | {avg} |
### Blockers Encountered
| Blocker Type | Count | Resolution |
|--------------|-------|------------|
| {type} | {N} | {how resolved} |
After writing the report:
ANALYSIS COMPLETE
=================
Analyzed: {N} sessions across {M} projects
Report: .claude/logs/ANALYSIS_REPORT.md
Key Findings:
- {top finding 1}
- {top finding 2}
- {top finding 3}
Cross-Project:
- {cross-project insight 1}
- {cross-project insight 2}
Recommended next steps:
1. Review .claude/logs/ANALYSIS_REPORT.md for full details
2. Consider creating skills for high-priority patterns
3. Update AGENTS.md with guidance for common questions
If ~/.claude/logs/ directory doesn't exist:
mkdir -p ~/.claude/logsIf session log file is malformed:
If transcript files are inaccessible:
If no patterns are found:
If ANALYSIS_REPORT.md cannot be written: