Cta
This skill should be used when the user asks to "analyze token usage", "check costs", "how much did I spend", "token analysis", "session analysis", "cta", or mentions Claude Code usage efficiency, token spending, or cost optimization. Routes to 6 workflow sub-skills: cta-health-check, cta-cost-audit, cta-anomaly-hunt, cta-project-review, cta-trend-watch, cta-usage-pattern. Also handles ambiguous intent with a quick overview. Do NOT trigger for: building token-related features, tokenizer/NLP work, or non-Claude cost analysis.From its SKILL.md
npx -y skills add li195111/claude-token-analyzer --skill ctaAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 15 stars15 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.6 KB, as published. Nobody here has run it
CTA — Claude Token Analyzer Router
Route token analysis requests to the appropriate workflow skill, or provide a quick overview for ambiguous intent.
Prerequisites
- sync_db when freshness matters — Execute
mcp__token-analyzer__sync_dbbefore workflows that read SQLite-backed aggregates (analyze_global,analyze_project,cost_report,anomaly_scan,trend_report) when the user asks for latest data. Historicalclassify_session_patternlookups can run directly against JSONL without waiting forsync_db. - Output language — Use 繁體中文 for prose. Keep English for technical terms (cache_hit_rate, session_id, subagent, tool names).
- Error handling — Report MCP tool errors to the user; never swallow silently. State explicitly when results are empty.
Intent Routing
| User Intent | Keywords | Route to |
|---|---|---|
| Quick status | 「看看」「總覽」「花了多少」「現在狀況」 | cta-health-check |
| Monthly cost | 「這個月」「成本報告」「月度」「預算」 | cta-cost-audit |
| Find anomalies | 「異常」「有問題嗎」「診斷」「排查」 | cta-anomaly-hunt |
| Project analysis | 「分析專案」「專案健檢」「subagent」「工具使用」 | cta-project-review |
| Trend forecast | 「趨勢」「在漲嗎」「預測」「燃率」 | cta-trend-watch |
| Usage pattern | 「使用模式」「pattern 分析」「harness 優化」「工作流建議」「sparkline」 | cta-usage-pattern |
Routing Behavior
- Clear intent — Invoke the matching sub-skill directly.
- Ambiguous intent (「幫我看看 token」「分析一下」) — If latest data matters, run sync_db → analyze_global → output a one-page summary (format below) → ask which direction to explore.
- Cross-domain — Sub-skills may route to each other mid-workflow.
Shared Output Format
| Element | Format |
|---|---|
| Currency | $X.XX USD |
| Percentage | One decimal (85.3%) |
| session_id | First 8 chars (a1b2c3d4) |
| Token count | Thousands separator (125,000) |
| Large output (>50K chars) | Write to ${TMPDIR:-/tmp}/cta-*.md, report path |
| Cache hit rate <70% | Mark with warning |
| Subagent ratio >20% | Mark with notice |
One-Page Summary Template (for ambiguous intent)
## CTA Overview
| Metric | Value |
|--------|-------|
| Total Sessions | X |
| Total Projects | X |
| Total Cost | $X.XX USD |
| Avg Cache Hit Rate | X.X% |
| Subagent Token Ratio | X.X% |
### Top 3 Projects (by cost)
1. project-name — $X.XX (N sessions)
### Top 3 Sessions (by cost)
1. a1b2c3d4 — $X.XX (project-name)
Follow with: 「要深入哪個方向?成本 / 異常 / 專案 / 趨勢」
Additional Resources
Reference Files
For complete MCP tool parameters, return types, and known caveats:
references/tool-reference.md— 8 MCP tool quick reference with parameter defaults, advanced tuning, and known pitfalls
Sub-skills reference this file at the absolute path ${CLAUDE_PLUGIN_ROOT}/skills/cta/references/tool-reference.md.