Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →複数のストレージレイヤー（ローカルファイル、MCP メモリ、ベクターストア、Git リポジトリ）にわたるナレッジベースの管理、取り込み、同期、検索。ユーザーが知識システム全体で保存・整理・同期・重複排除・検索を行いたい場合に使用します。
.claude/skills/affaan-m-knowledge-ops/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 202% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 50% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 266% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 89% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 37% | 0% |
管理一个多层知识系统，用于跨多个存储库进行知识的摄取、组织、同步和检索。
推荐使用实时工作区模型：
~/.claude/projects/*/memory/当需要捕获新知识时：
这是什么类型的知识？
检查此知识是否已存在：
写入适当的层级：
更新任何相关的索引或摘要文件。
定期将会话历史同步到知识库：
将重要的工作区配置和脚本镜像到知识库：
当信息影响活跃执行时：
将来自多个来源的知识汇集到一处：
# 短期：当前会话上下文
使用 TodoWrite 进行会话内任务追踪
# 中期：项目记忆文件
写入 ~/.claude/projects/*/memory/ 以实现跨会话回溯
# 长期：GitHub / Linear / 知识库
将活跃执行事实置于 GitHub + Linear
将持久化综合上下文置于知识库仓库
# 语义层：MCP 知识图谱
使用 mcp__memory__create_entities 创建永久结构化数据
使用 mcp__memory__create_relations 进行关系映射
使用 mcp__memory__add_observations 添加关于已知实体的新事实
使用 mcp__memory__search_nodes 查找已有知识在完成任何知识操作之前：
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→fail | 6,622 | 4,916 | -26% | 1 | 1 | 0% | 1,033 | 2,300 | +123% | 0 | 0 | — |
case-01 | fail→fail | 14,827 | 9,132 | -38% | 1 | 1 | 0% | 2,546 | 2,157 | -15% | 0 | 0 | — |
case-02 | fail→fail | 16,607 | 6,610 | -60% | 1 | 1 | 0% | 2,478 | 1,847 | -25% | 0 | 0 | — |
case-03 | fail→pass | 8,593 | 13,972 | +63% | 1 | 1 | 0% | 1,327 | 4,003 | +202% | 0 | 0 | — |
case-04 | fail→fail | 2,878 | 9,978 | +247% | 1 | 1 | 0% | 304 | 2,107 | +593% | 0 | 0 | — |
case-05 | fail→pass | 17,526 | 15,866 | -9% | 1 | 1 | 0% | 2,689 | 4,031 | +50% | 0 | 0 | — |
case-06 | fail→pass | 11,345 | 27,411 | +142% | 1 | 1 | 0% | 1,522 | 5,566 | +266% | 0 | 0 | — |
case-08 | fail→fail | 7,189 | 3,374 | -53% | 1 | 1 | 0% | 1,199 | 1,871 | +56% | 0 | 0 | — |
case-09 | fail→fail | 11,975 | 5,557 | -54% | 1 | 1 | 0% | 1,991 | 2,146 | +8% | 0 | 0 | — |
case-10 | fail→pass | 7,349 | 4,854 | -34% | 1 | 1 | 0% | 1,157 | 2,191 | +89% | 0 | 0 | — |
case-11 | fail→fail | 7,947 | 16,837 | +112% | 1 | 1 | 0% | 1,289 | 3,981 | +209% | 0 | 0 | — |
case-12 | pass→pass | 11,732 | 10,881 | -7% | 1 | 1 | 0% | 1,791 | 3,265 | +82% | 0 | 0 | — |
case-13 | fail→fail | 7,908 | 10,078 | +27% | 1 | 1 | 0% | 1,395 | 1,825 | +31% | 0 | 0 | — |
case-14 | fail→fail | 9,131 | 5,883 | -36% | 1 | 1 | 0% | 1,551 | 1,755 | +13% | 0 | 0 | — |
case-15 | fail→fail | 12,203 | 7,660 | -37% | 1 | 1 | 0% | 1,894 | 2,632 | +39% | 0 | 0 | — |
case-16 | fail→pass | 11,817 | 6,951 | -41% | 1 | 1 | 0% | 1,852 | 2,546 | +37% | 0 | 0 | — |
case-17 | fail→pass | 8,156 | 10,146 | +24% | 1 | 1 | 0% | 1,209 | 3,228 | +167% | 0 | 0 | — |
case-18 | fail→pass | 6,384 | 3,400 | -47% | 1 | 1 | 0% | 998 | 2,039 | +104% | 0 | 0 | — |
case-19 | pass→pass | 13,220 | 8,692 | -34% | 1 | 1 | 0% | 1,528 | 2,840 | +86% | 0 | 0 | — |
case-20 | pass→pass | 9,379 | 4,724 | -50% | 1 | 1 | 0% | 1,449 | 2,161 | +49% | 0 | 0 | — |
case-21 | pass→pass | 10,377 | 6,381 | -39% | 1 | 1 | 0% | 1,972 | 2,661 | +35% | 0 | 0 | — |
case-22 | pass→fail | 7,427 | 4,066 | -45% | 1 | 1 | 0% | 1,090 | 1,904 | +75% | 0 | 0 | — |
case-23 | pass→pass | 8,707 | 9,035 | +4% | 1 | 1 | 0% | 1,430 | 3,002 | +110% | 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. 23 cases were attempted, and 19 counted toward the lift figure. The other 4 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 +26 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 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.