Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →ECC ツール、エージェント、スキル、および実装のコスト監査を実施します。プロンプト入力トークンを分析して、計算効率を定量化します。
.claude/skills/affaan-m-ecc-tools-cost-audit/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 445% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 47% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 60% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 37% | 0% |
当用户怀疑 ECC Tools GitHub App 正在消耗成本、过度创建 PR、绕过使用限制，或将免费用户引导至付费分析路径时，使用此技能。
这是一个针对兄弟仓库 ECC-Tools 的聚焦操作者工作流。它不是通用的计费技能，也不是仓库范围的代码审查。
在相关情况下，将这些 ECC 原生技能拉入工作流：
autonomous-loops 用于跨 webhook、队列、计费和重试的有界多步骤审计agentic-engineering 用于将请求路径追踪为离散的、可证明的单元customer-billing-ops 当需要清晰分离仓库行为和客户影响计算时search-first 在发明辅助函数或重新实现仓库本地工具之前security-review 当涉及认证、使用限制、授权或密钥时verification-loop 用于证明重试安全性和精确的修复后状态tdd-workflow 当修复需要在 worker、路由器或计费路径中添加回归测试覆盖时ECC-Tools 中，并依赖于 webhook 处理器、队列 worker、使用预留、PR 创建逻辑或付费网关强制执行ECC-Tools 中工作，而非 everything-claude-codeECC-Toolssrc/index.* 或主入口点如果用户要求代码更改，请按以下顺序优先修复：
除非同一根本原因明显跨越多个文件，否则将修复范围限制在一到三个直接修复。
如果推送、PR 同步和手动审计都入队相同的任务，并且 worker 总是创建 PR，那么分析就等于 PR 垃圾信息。
如果在入口处检查使用量，但仅在 worker 中增加，则并发请求可能全部通过关卡并超出配额。
如果存在密钥时，免费的排队任务仍能路由到 Anthropic 或其他付费提供商，即使客户从未看到付费结果，这也是真实的支出泄漏。
如果 pull_request.synchronize、分支推送或评论触发的运行在应用拥有的分支上触发，则应用可以递归分析自己的输出。
如果系统可能消耗令牌，然后在 PR 创建、文件更新或分支冲突时失败，则是在消耗成本而不产生价值。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-20 | pass→pass | 11,212 | 6,824 | -39% | 1 | 1 | 0% | 1,669 | 2,527 | +51% | 0 | 0 | — |
case-21 | pass→pass | 15,588 | 8,025 | -49% | 1 | 1 | 0% | 2,137 | 2,666 | +25% | 0 | 0 | — |
case-01 | fail→fail | 10,374 | 11,035 | +6% | 1 | 1 | 0% | 297 | 1,889 | +536% | 0 | 0 | — |
case-02 | fail→pass | 9,115 | 27,666 | +204% | 1 | 1 | 0% | 1,052 | 5,733 | +445% | 0 | 0 | — |
case-03 | pass→pass | 26,111 | 26,248 | +1% | 1 | 1 | 0% | 2,973 | 4,649 | +56% | 0 | 0 | — |
case-04 | pass→fail | 15,558 | 6,012 | -61% | 1 | 1 | 0% | 3,084 | 1,775 | -42% | 0 | 0 | — |
case-05 | pass→fail | 20,522 | 7,054 | -66% | 1 | 1 | 0% | 4,264 | 2,098 | -51% | 0 | 0 | — |
case-06 | pass→fail | 7,996 | 5,057 | -37% | 1 | 1 | 0% | 1,426 | 1,784 | +25% | 0 | 0 | — |
case-07 | pass→fail | 13,346 | 6,285 | -53% | 1 | 1 | 0% | 2,033 | 1,782 | -12% | 0 | 0 | — |
case-08 | pass→pass | 10,796 | 5,811 | -46% | 1 | 1 | 0% | 1,574 | 2,454 | +56% | 0 | 0 | — |
case-09 | fail→pass | 16,646 | 14,552 | -13% | 1 | 1 | 0% | 2,561 | 3,764 | +47% | 0 | 0 | — |
case-10 | pass→pass | 10,368 | 2,500 | -76% | 1 | 1 | 0% | 1,930 | 1,962 | +2% | 0 | 0 | — |
case-11 | pass→pass | 12,167 | 10,006 | -18% | 1 | 1 | 0% | 1,924 | 3,052 | +59% | 0 | 0 | — |
case-12 | pass→pass | 9,585 | 9,141 | -5% | 1 | 1 | 0% | 1,630 | 2,873 | +76% | 0 | 0 | — |
case-13 | pass→fail | 13,925 | 3,853 | -72% | 1 | 1 | 0% | 2,376 | 1,959 | -18% | 0 | 0 | — |
case-14 | pass→pass | 13,333 | 9,847 | -26% | 1 | 1 | 0% | 2,052 | 3,193 | +56% | 0 | 0 | — |
case-15 | pass→pass | 7,518 | 3,084 | -59% | 1 | 1 | 0% | 1,199 | 2,018 | +68% | 0 | 0 | — |
case-16 | fail→pass | 16,447 | 3,815 | -77% | 1 | 1 | 0% | 1,806 | 2,078 | +15% | 0 | 0 | — |
case-17 | pass→pass | 10,163 | 3,342 | -67% | 1 | 1 | 0% | 1,498 | 2,100 | +40% | 0 | 0 | — |
case-18 | fail→pass | 13,130 | 13,229 | +1% | 1 | 1 | 0% | 2,056 | 3,296 | +60% | 0 | 0 | — |
case-19 | fail→pass | 12,289 | 6,707 | -45% | 1 | 1 | 0% | 1,870 | 2,568 | +37% | 0 | 0 | — |
case-22 | fail→pass | 9,785 | 6,489 | -34% | 1 | 1 | 0% | 1,555 | 2,539 | +63% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 17 counted toward the lift figure. The other 5 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +5 percentage points is the difference between those two pass rates over the 17 comparable cases. 5 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.