curator(2026-09-06): 技能库治理全部执行 — 补34个version/修8处name对齐/排除node_modules假技能/归档blocked-page-recovery+plan重复
This commit is contained in:
小唯 A06
parent
5ce7390c2c
commit
cbe560ce98
|
|
@ -70,6 +70,7 @@ def load_all_skills():
|
|||
"""遍历 skills/ 下所有子目录，读取 SKILL.md frontmatter"""
|
||||
skills = []
|
||||
for root, dirs, files in os.walk(SKILLS_DIR):
|
||||
dirs[:] = [d for d in dirs if d != "node_modules"] # 排除 node_modules 假技能 (2026-09-06)
|
||||
if ".archive" in root or ".quarantine" in root or "_archive" in root or "_quarantine" in root:
|
||||
continue
|
||||
if "SKILL.md" in files:
|
||||
|
|
|
|||
|
|
@ -39,6 +39,7 @@ def scan_skills():
|
|||
|
||||
# 先扫 active skills
|
||||
for root, dirs, files in os.walk(SKILLS_DIR):
|
||||
dirs[:] = [d for d in dirs if d != "node_modules"] # 排除 node_modules 假技能 (2026-09-06)
|
||||
if "SKILL.md" in files:
|
||||
path = os.path.join(root, "SKILL.md")
|
||||
meta = parse_skill(path)
|
||||
|
|
|
|||
|
|
@ -0,0 +1,344 @@
|
|||
# 多技能组合手册 — 小唯 A06（技能链 × 任务画像）
|
||||
|
||||
> 版本：1.0.0 | 2026-09-06 | kanban t_52c49737 · prof-b
|
||||
> 一句话：**单技能是一把刀，技能链是一条流水线。** 本手册把 301 个 active 技能编排成「任务画像 → 技能链」组合，覆盖我们真实高频场景。
|
||||
> 技能名全部在 `~/.hermes/skills/` 磁盘真实存在（含 1 个符号链接 nuyoah-xiezhen-prompt → .agents/skills），可脚本验证。
|
||||
> 用法：接到任务先对画像，命中就按链加载；每条链 2-4 个技能，跨 category 才叫组合。
|
||||
|
||||
---
|
||||
|
||||
## 第一部分 通用技能层（元技能）— 横切一切专业任务
|
||||
|
||||
> 牧尘 2026-09-06 指令：通用技能 = 横切所有专业任务的「思维/质量前提层」，独立成层。
|
||||
> 本部分只列**已存在**的元技能（不编造）；执行任何专业任务时，按 §1.2 的叠加模式套用。
|
||||
|
||||
### 1.1 通用技能清单（按思维动作分组，全部真实存在）
|
||||
|
||||
| 思维动作 | 通用技能（元技能） | 真实路径 | 一句话 |
|
||||
|---|---|---|---|
|
||||
| **规划** | plan | software-development/plan | 写可执行 markdown 计划（Plan mode；顶层 LaunchDarkly 外部 plan 已归档 .archive/plan-launchdarkly） |
|
||||
| **规划** | persistent-planning | persistent-planning/ | Manus 式持久化文件规划 |
|
||||
| **规划** | writing-plans | software-development/writing-plans | 写可执行实现计划（bite-size 任务） |
|
||||
| **规划** | kanban-orchestrator | devops/kanban-orchestrator | 分解 + 反诱惑规则 |
|
||||
| **规划** | task-breakdown | projectManagerBreakout/skills/task-breakdown | 大模糊问题拆成精确任务树 |
|
||||
| **规划** | sprint-planning | projectManagerBreakout/skills/sprint-planning | 排期/负载均衡 |
|
||||
| **规划** | dependency-analysis | projectManagerBreakout/skills/dependency-analysis | 依赖图/瓶颈检测 |
|
||||
| **三思/行动前验证** | systematic-debugging | software-development/systematic-debugging | 4 阶段根因调试（先懂后修） |
|
||||
| **三思/行动前验证** | requesting-code-review | software-development/requesting-code-review | 提交前审查（安全/质量门/自动修） |
|
||||
| **三思/行动前验证** | karpathy-code-discipline | karpathy-code-discipline/ | 代码纪律（最小化、诚实） |
|
||||
| **三思/行动前验证** | lazy-senior-dev | software-development/lazy-senior-dev | 7 阶梯最小化决策 |
|
||||
| **三思/行动前验证** | agent-feedback-control | autonomous-ai-agents/agent-feedback-control | PEV 反馈控制闭环 |
|
||||
| **反思/复盘** | retro | product-management/retro | 结构化复盘 |
|
||||
| **反思/复盘** | daily-recap | productivity/daily-recap | 每日复盘报告 |
|
||||
| **反思/复盘** | weekly-review-planning | productivity/weekly-review-planning | 每周重置承诺/停滞/下周 |
|
||||
| **反问/批判质疑** | strategy-red-team | product-management/strategy-red-team | 攻击方案假设（红队） |
|
||||
| **反问/批判质疑** | pre-mortem | product-management/pre-mortem | 事前预演失败 |
|
||||
| **反问/批判质疑** | merge-reconciler | autonomous-ai-agents/merge-reconciler | 中立第三方解决分歧 |
|
||||
| **反问/批判质疑** | intended-vs-implemented | product-management/intended-vs-implemented | 找意图与实现的差 |
|
||||
| **验收/完整性检查** | kanban-review-gate | autonomous-ai-agents/kanban-review-gate | PDCA 验收门禁（done≠完成） |
|
||||
| **验收/完整性检查** | sdlc-review | devops/sdlc-review | 评审 Kanban 交接/路由验证 |
|
||||
| **验收/完整性检查** | skill-curator-workflow | skill-curator-workflow/ | 技能 curator 审计流程 |
|
||||
| **验收/完整性检查** | hermes-self-improvement | autonomous-ai-agents/hermes-self-improvement | 技能生命周期/质量标尺/删除规范 |
|
||||
| **验收/完整性检查** | shipping-artifacts | product-management/shipping-artifacts | 交付物持久化文档集 |
|
||||
| **验收/完整性检查** | dogfood | dogfood/ | 探索性 QA 找 bug |
|
||||
| **验收/完整性检查** | website-ux-audit | productivity/website-ux-audit | 全站 UX/代码质量审计 |
|
||||
| **通用流程入口** | team-composer | autonomous-ai-agents/team-composer | 144 skill + 184 ao 角色组队 |
|
||||
| **通用流程入口** | project-toolbox | knowledge/project-toolbox | 项目工具箱总索引 |
|
||||
| **记忆质量** | memory-routing-contract | knowledge/memory-routing-contract | 记忆路由矩阵 |
|
||||
| **记忆质量** | memory-governance | knowledge/memory-governance | 注入预算/淘汰/分层 |
|
||||
| **记忆质量** | memory-verification | zhiyi/memory-verification | 写入前事实校验 |
|
||||
|
||||
> 注：`software-development/plan` 是小唯自己的 Plan mode 元技能（写 .hermes/plans/ 可执行计划）。曾被误当作 LaunchDarkly 外部套件（顶层 `plan/` 0.2.0，2026-09-06 已归档至 `.archive/plan-launchdarkly`）——真正做规划走 `persistent-planning` + `writing-plans` + `kanban-orchestrator`。
|
||||
|
||||
### 1.2 专业技能 × 通用技能的叠加模式（万能模板）
|
||||
|
||||
```
|
||||
一条完整专业任务 = 通用层(横切) × 专业层(领域)
|
||||
|
||||
① 规划层 persistent-planning / writing-plans / pre-mortem(高风险时) / task-breakdown
|
||||
② 执行层 专业技能(该任务领域技能，如 KOCR/织忆/ComfyUI/公文/调研)
|
||||
③ 三思/质疑 systematic-debugging(出错时) / strategy-red-team(方案重大时) / agent-feedback-control(迭代反馈)
|
||||
④ 验收证据 kanban-review-gate(PDCA) / requesting-code-review(代码) / skill-curator-workflow(技能)
|
||||
⑤ 复盘沉淀 retro / hermes-self-improvement(把模式存成技能) / daily-recap
|
||||
```
|
||||
|
||||
实例：牧尘发来一批凭证 PDF →
|
||||
① 规划：先想清批次/模板/合并规则（persistent-planning 落 /tmp 计划，防 gateway 重启失忆）
|
||||
② 执行：accounting-voucher-ocr（KOCR v8 单 PDF → 金蝶 K3 Excel）
|
||||
③ 三思：产出自检三关（摘要真实/科目干净/借贷平衡）——验收前置
|
||||
④ 验收：kanban-review-gate 过「运行证据」；发牧尘前自审
|
||||
⑤ 沉淀：新坑 → patch accounting-voucher-ocr skill
|
||||
|
||||
---
|
||||
|
||||
## 第二部分 任务画像 × 技能链（18 条高频画像）
|
||||
|
||||
> 规则：技能链按执行顺序写；每条注明各技能的真实 category；为什么单技能不够；组合后的 1+1>2。
|
||||
> 命名 = 目录名（skills_list 可用名）。
|
||||
|
||||
---
|
||||
|
||||
### C1. 竞品/行业调研报告（要交付一份可引用的分析）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | competitor-news-monitor（持续盯目标公司新闻，带引用摘要） | research |
|
||||
| 2 | web-content-extraction / scrapling（crawl4ai/scrapy/stealth 抓正文） | productivity / research |
|
||||
| 3 | grounded-citations（每条论断钉住可验证来源） | research |
|
||||
| 4 | team-composer（需要时并行多角色：研究员→分析师→摘要师） | autonomous-ai-agents |
|
||||
|
||||
- 为什么单技能不够：competitor-news-monitor 给「发生了什么」但抓不到深页正文；光抓取没有引用纪律 = 报告不可信；没有编排 = 所有维度自己串行跑。
|
||||
- 1+1>2：监控器喂素材 → 抓取补全文 → grounded-citations 强制带出处 → team-composer 并行拆维度，产出从「资讯堆」变成「可验证调研报告」。
|
||||
|
||||
---
|
||||
|
||||
### C2. 正式公文写作（请示/通知/函/总结，要合规+像人写的）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | gongwen-guobiao（党政公文格式国标） | top-level(gongwen 套件) |
|
||||
| 2 | gongwen-wenxi（六层算法：字词→句→段→结构→高观点） | top-level(gongwen 套件) |
|
||||
| 3 | gongwen-draft（起草/修订/润色/审查/导出） | top-level(gongwen 套件) |
|
||||
| 4 | gongwen-renmin（润色人性化，去 AI 腔） | top-level(gongwen 套件) |
|
||||
|
||||
- 为什么单技能不够：guobiao 管格式不管文采；wenxi 管算法但未必对齐国标；renmin 是最后人味关。只用一个 = 格式对但像模板，或文采好但格式错。
|
||||
- 1+1>2：guobiao 锁边界 → wenxi 组织内容 → draft 起草 → renmin 去机械感，产出「合规 + 专业 + 像人」三层都过的公文。（同族技能串联，顺序即流水线。）
|
||||
|
||||
---
|
||||
|
||||
### C3. AI 出图（文生图 → 验收 → 迭代）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | prompt-engineering（提示词工程总入口，先路由分发） | palace-prompts |
|
||||
| 2 | prompt-to-image-generation（提示词→Agnes API 出图实战链路） | creative |
|
||||
| 3 | ai-media-generation（Agnes API 生图/生视频多风格预设） | creative |
|
||||
| 4 | nuyoah-image-reverse-prompt（拿参考图反向拆提示词 / 验收一致性） | top-level |
|
||||
|
||||
- 为什么单技能不够：有提示词没 API 链路 = 出不了图；有 API 没提示词工程 = 图不对；出完不反向验收 = 和参考图偏差看不出。
|
||||
- 1+1>2：prompt-engineering 定风格策略 → prompt-to-image-generation 跑通出图 → 不满意时用 nuyoah-image-reverse-prompt 反推参考图提示词 → 回到 prompt 层迭代，形成「文字→图→文字→图」闭环。
|
||||
|
||||
---
|
||||
|
||||
### C4. AI 写真工作流（人像参考图 → 多场景写真）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | nuyoah-xiezhen-prompt（写真/editorial 中文提示词生成改写） | top-level(symlink) |
|
||||
| 2 | ai-portrait-workflow（参考图 → 多场景写真标准工作流） | creative |
|
||||
| 3 | comfyui-portrait-optimization（ComfyUI 真人肖像生成优化技巧） | creative |
|
||||
| 4 | comfyui-rtx3050（本地 4GB 显存跑 SD 1.5 的坑/选型） | creative |
|
||||
|
||||
- 为什么单技能不够：写真成败在提示词(1)+管线(2)+本地出图参数(3/4)。只用提示词 → 云端图贵且风格飘；只调 ComfyUI → 不知道怎么写人像提示词。
|
||||
- 1+1>2：nuyoah-xiezhen-prompt 产出专业人像提示词 → ai-portrait-workflow 定多场景管线 → 本地用 comfyui-rtx3050 参数出图 + comfyui-portrait-optimization 调脸，成本从云端 API 降到本地 GPU。
|
||||
|
||||
---
|
||||
|
||||
### C5. PPT / 演示文稿（要一页页好看、可汇报）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | design-system-discipline（三层 token 视觉纪律，避免风格漂移） | top-level |
|
||||
| 2 | guizang-ppt-skill（横向翻页网页 PPT，WebGL 背景+讲稿备注） | top-level |
|
||||
| 3 | frontend-slides（动画丰富 HTML 演示） | top-level |
|
||||
| 4 | dashi-ppt（主题化 PPT 生成，可离线打开） | top-level |
|
||||
|
||||
- 为什么单技能不够：dashi-ppt/guizang/frontend-slides 三个都是「生成器」但视觉基因不同；不先定 design-system 纪律，三个工具混用 = 每页一个风格。
|
||||
- 1+1>2：design-system-discipline 定 token（颜色/字号/间距）→ 选 guizang（正式汇报）或 frontend-slides（炫技场）生成 → 快速出「一套视觉一致」的网页 PPT，而不是单工具默认模板拼盘。
|
||||
|
||||
---
|
||||
|
||||
### C6. 短剧/漫剧/视频创作（剧本 → 分镜 → 素材 → 配音）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | drama-script-writing（短剧/漫剧编剧方法论） | content |
|
||||
| 2 | ai-video-storyboard-prompts（分镜/运镜提示词工程） | creative |
|
||||
| 3 | ai-media-generation（Agnes 出图/出视频） | creative |
|
||||
| 4 | tts-stt-tools 或 mimo-tts-pipeline（配音 TTS + opus 转码） | media |
|
||||
|
||||
- 为什么单技能不够：编剧给故事不给镜头；分镜给画面描述不给素材；素材生成了没人配音 = 片子不完整。四个环节环环依赖。
|
||||
- 1+1>2：剧本 → 每场戏转分镜提示词 → Agnes 批量出素材 → TTS 配音合成，形成一条「文字→镜头→像素→声音」的短视频流水线（media category 补上最后一块）。
|
||||
|
||||
---
|
||||
|
||||
### C7. 股票投研（选股 → 回测 → 每日信号）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | stock-research（四维选股/回测对比/每日信号） | research |
|
||||
| 2 | xlsx（把回测结果落成表格/对比） | productivity |
|
||||
| 3 | provider-tiering（信号跑在免费模型上省钱） | devops |
|
||||
|
||||
- 为什么单技能不够：stock-research 给策略和信号；但结果要给人看要落 xlsx；且「每日信号」如果跑付费模型成本高。
|
||||
- 1+1>2：stock-research 出信号 → xlsx 导出对比表 → cron 定时 + provider-tiering 选免费模型，形成低成本每日投研流水线（实际已有 stock_daily_signal_paper.sh 佐证该模式）。
|
||||
|
||||
---
|
||||
|
||||
### C8. 自媒体运营（公众号/小红书 0→1 + 多平台分发）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | marketing-disciplines（营销/SEO/CRO 方法论 + 科研工具索引） | top-level |
|
||||
| 2 | multi-industry-prompts（70+ 行业文案模板） | content |
|
||||
| 3 | social-media-ops（公众号+小红书从 0 到 1、内容创作分发） | social-media |
|
||||
| 4 | feishu（做好的内容/协作发飞书） | social-media |
|
||||
|
||||
- 为什么单技能不够：营销方法论 ≠ 会写文案；文案模板 ≠ 懂平台调性；写完全靠手工发 = 不可持续。
|
||||
- 1+1>2：marketing-disciplines 定选题/转化框架 → multi-industry-prompts 出对应行业文案 → social-media-ops 按平台规则排版分发 → feishu 推送预览，把「写一篇」变成「平台运营流水线」。
|
||||
|
||||
---
|
||||
|
||||
### C9. 会议/访谈 → 行动项（要纪要+负责人+截止）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | teams-meeting-pipeline（Teams 会议记录管线） | productivity |
|
||||
| 2 | summarize-meeting（结构化纪要） | top-level |
|
||||
| 3 | meeting-action-items（决策/负责人/DDL 提取） | productivity |
|
||||
| 4 | weekly-review-planning（落进每周承诺/下周计划） | productivity |
|
||||
|
||||
- 为什么单技能不够：光纪要没行动项 = 听完就忘；光提行动项不进周计划 = 无人跟进。链路缺一不可。
|
||||
- 1+1>2：会议转写 → summarize-meeting 成结构化纪要 → meeting-action-items 抽出「谁在何时前做什么」→ 汇入 weekly-review-planning，会议从「记录」变「可执行承诺」。（document-to-action-items 可补合同/文档类提取。）
|
||||
|
||||
---
|
||||
|
||||
### C10. 网页抓取/反封锁（正文抓不到 / 403 / 墙）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | blocked-page-recovery（403/429/paywall/WAF 回退；research 版为 canonical，web 副本已归档） | research |
|
||||
| 2 | scrapling（stealth 浏览 + Cloudflare 绕过） | research |
|
||||
| 3 | web-content-extraction（crawl4ai/scrapy/camofox/pymupdf 统一工具链） | productivity |
|
||||
| 4 | firecrawl（网页→Markdown 一键） | productivity |
|
||||
|
||||
- 为什么单技能不够：firecrawl 能抓普通页但遇墙就挂；blocked-page-recovery 给策略不给你工具命令；scrapling 专破 Cloudflare 但输出糙。
|
||||
- 1+1>2：firecrawl 先试 → 403 时 blocked-page-recovery 选回退路线 → scrapling stealth 硬抓 → 全文交给 web-content-extraction 统一清洗/转 md，形成「普通→硬核→清洗」三级抓取管线。
|
||||
|
||||
---
|
||||
|
||||
### C11. 记忆系统维护（织忆/注入记忆 出问题或要治理）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | memory-routing-contract（拿不准用哪套记忆/矛盾/自纠时） | knowledge |
|
||||
| 2 | zhiyi（织忆 API/健康/运维规范，4 组件健康检查） | zhiyi |
|
||||
| 3 | memory-verification（写入前检测矛盾/猜测/脱敏误判） | zhiyi |
|
||||
| 4 | memory-governance（注入预算 + 淘汰 + 存储分层） | knowledge |
|
||||
|
||||
- 为什么单技能不够：zhiyi 管织忆服务本身，不管「该不该写这条记忆」；memory-governance 管预算但不懂织忆 API；没有 memory-verification = 脏记忆一直写进去。
|
||||
- 1+1>2：memory-routing-contract 先裁决走哪套记忆 → zhiyi 拉真实健康 → 写入前 memory-verification 校验 → memory-governance 控制总量，形成「路由→健康→防污染→预算」记忆治理闭环。
|
||||
|
||||
---
|
||||
|
||||
### C12. 技能库治理（curator 周报 / 技能太多要理）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | skill-curator-workflow（curator 报告→人工判定→清理建议） | top-level |
|
||||
| 2 | hermes-self-improvement（技能生命周期/质量标尺/删除规范） | autonomous-ai-agents |
|
||||
| 3 | skill-library-porting（外部技能库批量移植） | knowledge |
|
||||
| 4 | project-toolbox（判断当前任务用哪个已落地项目） | knowledge |
|
||||
|
||||
- 为什么单技能不够：curator 只出机械分档会误判（多次教训）；self-improvement 管质量但不产报告；porting 只管导入。
|
||||
- 1+1>2：skill-curator-workflow 出本地真相报告 → hermes-self-improvement 的质量标尺 + 「删前验证」铁律兜底 → 需要恢复/移植时 skill-library-porting 走正规路径，技能治理从「瞎删」到「有门禁」。
|
||||
|
||||
---
|
||||
|
||||
### C13. 代码质量门禁（提交前 / PR 审查 / 守护进程）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | karpathy-code-discipline（写/审代码的诚实纪律） | top-level |
|
||||
| 2 | requesting-code-review（提交前安全扫描/质量门/自动修） | software-development |
|
||||
| 3 | github-code-review（PR 审查：diff/inline/gh/REST） | github |
|
||||
| 4 | daemon-code-quality（daemon.py/soulful_core/hermes 插件专项审查） | autonomous-ai-agents |
|
||||
|
||||
- 为什么单技能不够：requesting-code-review 是通用门；github-code-review 管 PR 流程；但小唯守护进程有专属坑（systemd/pid 存活/时区）——通用审查发现不了。
|
||||
- 1+1>2：karpathy 纪律打底 → requesting-code-review 本地自检 → github-code-review 走 PR 流程 → daemon-code-quality 专项扫守护进程，代码在「通用+专项」双保险下才放行。
|
||||
|
||||
---
|
||||
|
||||
### C14. 数据库/存储损坏恢复（sqlite/lancedb/state.db 报错）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | sqlite-db-corruption-recovery（SQLite 11 种损坏模式修复） | devops |
|
||||
| 2 | lancedb-corruption-recovery（LanceDB 0 字节/向量丢失） | devops |
|
||||
| 3 | cron-watchdog-hygiene（事故后审计 cron/watchdog，去重去自愈杀伤） | devops |
|
||||
| 4 | self-healing-infrastructure（监控/回滚/自进化大框架） | devops |
|
||||
|
||||
- 为什么单技能不够：sqlite 修法 ≠ lancedb 修法 ≠ 事后 cron 审计；只修库不审 cron = 下次自愈脚本又杀库。
|
||||
- 1+1>2：先按损坏类型进对应恢复技能 → 修完用 cron-watchdog-hygiene 排查「是不是 watchdog 误杀」→ self-healing-infrastructure 把防再发写进监控，单次救火变成系统免疫。
|
||||
|
||||
---
|
||||
|
||||
### C15. 本地 LLM / GPU 推理部署（llama.cpp / 显存规划）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | local-ai-infrastructure（需求/架构/硬件/模型选型） | mlops |
|
||||
| 2 | local-llm-capacity-planning（小内存机选常驻模型：host-RSS/reasoning 坑） | mlops |
|
||||
| 3 | llama-cpp（GGUF 本地推理 + HF 模型发现） | mlops |
|
||||
| 4 | local-gpu-inference（Vulkan/ONNX CUDA 调优） | mlops |
|
||||
|
||||
- 为什么单技能不够：选型不容量规划 = 4GB 显存跑 7B 直接爆；容量对了不会 llama.cpp 命令 = 跑不起来；跑起来不会 GPU 调优 = 9t/s 难受。
|
||||
- 1+1>2：local-ai-infrastructure 定架构 → local-llm-capacity-planning 按 16GB/4GB 硬件选 Q4 模型 → llama-cpp 部署 → local-gpu-inference 切 Vulkan/调参，复现「bge 走 CPU、llama 全 GPU → 14-43 t/s」的笔记本最优解。
|
||||
|
||||
---
|
||||
|
||||
### C16. Docker/服务部署 + 内网穿透（容器化上公网）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | docker / docker-management（构建/运行/编排） | devops |
|
||||
| 2 | nginx（反代/SSL/负载均衡） | devops |
|
||||
| 3 | self-hosted-tunneling（FRP：frps 云 + frpc 家服，公网暴露） | devops |
|
||||
| 4 | windows-home-server-ops（家服 Windows/ZSB GPU 部署运维） | devops |
|
||||
|
||||
- 为什么单技能不够：docker 管容器不管入口；nginx 管入口不暴露公网；tunneling 只给隧道；家服那台 Windows 有自己坑（frpc.exe 闪退、DirectML）。
|
||||
- 1+1>2：docker 起服务 → nginx 反代/SSL → self-hosted-tunneling 走 frpc（iStoreOS 24h 跑，避开 Win frpc.exe 坑）→ windows-home-server-ops 按家服真实拓扑运维，形成「容器→入口→公网→硬件」完整上线路。
|
||||
|
||||
---
|
||||
|
||||
### C17. 语音管线（STT 转写 + TTS 合成 + 转码发送）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | whisper-stt（本地 faster-whisper 离线转写） | top-level |
|
||||
| 2 | tts-stt-tools（edge-tts + GLM-ASR 免本地模型方案） | media |
|
||||
| 3 | mimo-tts-pipeline（MiMo TTS wav→opus + ffmpeg 参数） | media |
|
||||
| 4 | feishu（发语音消息到飞书） | social-media |
|
||||
|
||||
- 为什么单技能不够：whisper 要下模型本地跑；tts-stt-tools 走云端 API 但转码不在行；mimo 强在音质但没平台发送。
|
||||
- 1+1>2：按资源选 whisper（离线）或 tts-stt-tools（云端）→ 合成后 mimo-tts-pipeline 转 opus（飞书不吃 wav）→ feishu 发语音，形成「转写/合成→转码→发送」闭环。
|
||||
|
||||
---
|
||||
|
||||
### C18. 任务委派/多智能体编排（大目标 → 拆 → 并行 → 验收）
|
||||
|
||||
| 步骤 | 技能 | category |
|
||||
|---|---|---|
|
||||
| 1 | kanban-routing / kanban-router（先判断路由：research/dsh/npc/opencode/delegate） | top-level |
|
||||
| 2 | team-composer（按任务类型组队：144 skill + 184 ao 角色） | autonomous-ai-agents |
|
||||
| 3 | ao-orchestrator（ao compose 多角色 DAG 执行） | autonomous-ai-agents |
|
||||
| 4 | subagent-driven-development（delegate_task 子代理两段式审查执行） | software-development |
|
||||
|
||||
- 为什么单技能不够：kanban-routing 只告诉你「派给谁」，不给怎么组队；team-composer 给组队不给执行引擎；ao 给引擎但复杂任务要显式 YAML；subagent 给子代理但需要父进程验收。
|
||||
- 1+1>2：kanban-routing 定路由 → team-composer 定角色组合 → ao compose 生成 DAG 并行 → 关键代码走 subagent-driven-development 两段式审查，把「模糊大目标」变「多角色并行流水线 + 验收」。（全链路实际已被 kanban 分身/ao 阵法验证。）
|
||||
|
||||
---
|
||||
|
||||
## 附 A：验证命令（每个技能名都在磁盘存在）
|
||||
|
||||
```bash
|
||||
cd ~/.hermes/skills && for n in <技能名>; do find . -maxdepth 4 -type d -name "$n" -not -path '*/node_modules/*' | head -1; done
|
||||
# nuyoah-xiezhen-prompt 是符号链接 → ls -la ~/.hermes/skills/nuyoah-xiezhen-prompt
|
||||
```
|
||||
|
||||
## 附 B：组合使用提示
|
||||
|
||||
1. **同一任务多个画像命中** → 链式叠加（如 C9 → C13：会议提出代码任务 → 代码质量门禁）。
|
||||
2. **每链第 1 个通常是「入口/策略」技能**（prompt-engineering / skill-curator-workflow / memory-routing-contract 类），起路由作用。
|
||||
3. **通用技能层优先于专业链内嵌**：专业链只管领域执行，通用层（规划/验收/复盘）按 §1.2 在外面套，不要写进每一条（避免重复）。
|
||||
4. 组合里技能顺序 = 数据流方向；顺序错了效果打折（例如先抓取后定引用纪律 = 白抓）。
|
||||
|
|
@ -0,0 +1,323 @@
|
|||
# 技能库盘点报告 2026-09-06
|
||||
|
||||
> 生成：2026-09-06 10:52（kanban t_52c49737 · prof-b，run 182）
|
||||
> 校验：2026-09-06 10:55 run 183 重跑 skill-curator.py + skill-manager.py scan —— 数字一致（303 active / A0 B80 C113 D110 / T0 51 T1 107 T2 113 T3 32 / 20 重叠 / 32 归档候选），本报告与本地 JSON 一致。
|
||||
> 数据源：skill-curator.py --report（10:41 与 10:55 两跑一致）+ skill-manager.py scan + 磁盘 SKILL.md 扫描 + durable 引用 grep
|
||||
> 性质：**只读盘点**。零删除、零归档动作；全部建议等牧尘批准后执行。
|
||||
|
||||
---
|
||||
|
||||
## 1. 当前技能库总览
|
||||
|
||||
| 指标 | 数值 | 说明 |
|
||||
|---|---|---|
|
||||
| health 报告 total | 419 | active 303 + archived 116 |
|
||||
| curator 报告 total | 303 | active（含 2 个 node_modules 假技能 skill/trace） |
|
||||
| 磁盘真实 active | 301 | find ~/.hermes/skills -name SKILL.md，排除 .archive/.quarantine/node_modules |
|
||||
| 等级分布 | A 0 / B 80 / C 113 / D 110 | health computed score /10 |
|
||||
| Tier 分布 | T0 51 / T1 107 / T2 113 / T3 32 | curator |
|
||||
| 分类（health 按目录） | 87 类 | 见 §3 |
|
||||
| 缺 version（磁盘级） | 36 | 见 §5 |
|
||||
| needs_attention | 161 | 多为缺 tags/refs/related，非删除理由 |
|
||||
|
||||
### 扫描器差异说明（诚实标注）
|
||||
|
||||
- **curator 用 frontmatter name 做主键，health 用目录名做主键** → 同一技能两种名字出现在清单（如 curator 报 `ideation` = health 的 `creative-ideation`；curator 报 `project-discovery` = health 的 `github-project-discovery`）。
|
||||
- **health 会扫到 `dashi-ppt/project/node_modules/playwright-core/` 里 2 个假技能**（`skill`、`trace`，Playwright 自带 SKILL.md）→ 所以 health active=303 但磁盘真实只有 301。
|
||||
- 本报告以**磁盘 ground truth（301）+ curator tier3 + health grade D 的并集**为准。
|
||||
|
||||
---
|
||||
|
||||
## 2. 通用技能（元技能）vs 专业技能 两分类
|
||||
|
||||
> 牧尘 2026-09-06 增补指令：把横切所有专业任务的「思维/质量前提层」独立出来。
|
||||
|
||||
### 2.1 判定标准
|
||||
|
||||
| 类别 | 特征 | 判定方法 |
|
||||
|---|---|---|
|
||||
| **通用技能（元技能）** | 横切多个专业域；质量前提/思维框架/流程规范；与具体业务无关 | 名字含 plan/review/audit/debug/retro/metric/gate/principle/discipline/quality/routing/curator 等；内容描述是方法论文而非某工具某平台 |
|
||||
| **专业技能** | 绑定具体工具/平台/领域（ComfyUI、飞书、金蝶、织忆、NewAPI…） | 描述含特定产品/API/库名 |
|
||||
|
||||
### 2.2 划分出的通用技能清单（36 个，已逐项核对存在）
|
||||
|
||||
| 通用技能 | 路径 | 元技能定位 |
|
||||
|---|---|---|
|
||||
| plan | software-development/plan, plan/ | 计划模式（**重复 2 份**，见 §6） |
|
||||
| persistent-planning | persistent-planning/ | 持久化文件规划 |
|
||||
| writing-plans | software-development/writing-plans | 写可执行实现计划 |
|
||||
| systematic-debugging | software-development/systematic-debugging | 4 阶段根因调试 |
|
||||
| kanban-review-gate | autonomous-ai-agents/kanban-review-gate | 验收门禁（PDCA） |
|
||||
| kanban-orchestrator | devops/kanban-orchestrator | 分解 + 反诱惑规则 |
|
||||
| kanban-worker | devops/kanban-worker | 分身执行规范 |
|
||||
| kanban-routing | kanban-routing/ | 任务路由 |
|
||||
| kanban-router | kanban-router/ | 委派路由 |
|
||||
| skill-curator-workflow | skill-curator-workflow/ | curator 审计流程 |
|
||||
| hermes-self-improvement | autonomous-ai-agents/hermes-self-improvement | 技能生命周期规范 |
|
||||
| agent-feedback-control | autonomous-ai-agents/agent-feedback-control | PEV 反馈控制闭环 |
|
||||
| daemon-code-quality | autonomous-ai-agents/daemon-code-quality | 守护进程代码质量审查 |
|
||||
| pre-mortem | product-management/pre-mortem | 事前风险预演 |
|
||||
| strategy-red-team | product-management/strategy-red-team | 批判质疑（红队） |
|
||||
| retro | product-management/retro | 复盘回顾 |
|
||||
| sdlc-review | devops/sdlc-review | 评审 Kanban 交接 |
|
||||
| requesting-code-review | software-development/requesting-code-review | 提交前审查 |
|
||||
| subagent-driven-development | software-development/subagent-driven-development | 子代理执行计划 |
|
||||
| idea-superpowers-suite | idea-superpowers-suite/ | 点子→文档全流程 |
|
||||
| project-toolbox | knowledge/project-toolbox | 项目工具箱索引 |
|
||||
| memory-routing-contract | knowledge/memory-routing-contract | 记忆路由矩阵 |
|
||||
| memory-governance | knowledge/memory-governance | 记忆预算治理 |
|
||||
| memory-verification | zhiyi/memory-verification | 记忆事实校验 |
|
||||
| skill-library-porting | knowledge/skill-library-porting | 外部技能库移植 |
|
||||
| karpathy-code-discipline | karpathy-code-discipline/ | 代码纪律（最小化） |
|
||||
| lazy-senior-dev | software-development/lazy-senior-dev | 最小化决策 |
|
||||
| design-system-discipline | design-system-discipline/ | UI 设计系统纪律 |
|
||||
| grammar-check | product-management/grammar-check | 语法/逻辑检查 |
|
||||
| summarize-interview | product-management/summarize-interview | 访谈结构化总结 |
|
||||
| summarize-meeting | summarize-meeting/ | 会议纪要结构化 |
|
||||
| meeting-action-items | productivity/meeting-action-items | 会议行动项提取 |
|
||||
| document-to-action-items | productivity/document-to-action-items | 文档义务/截止提取 |
|
||||
| weekly-review-planning | productivity/weekly-review-planning | 每周重置规划 |
|
||||
| daily-recap | productivity/daily-recap | 每日复盘 |
|
||||
| team-composer | autonomous-ai-agents/team-composer | 团队作战手册（144 skill 组合） |
|
||||
|
||||
> 注：`ao-orchestrator` 偏工具链（ao 引擎），`so-team-workflow` 偏执行编排，归类为「协作专业技能」。`agent-feedback-control`/`kanban-review-gate`/`skill-curator-workflow` 是通用技能层里与「质量前提」最相关的 3 个。
|
||||
|
||||
### 2.3 通用技能 × 专业任务的叠加模式（给 §7 组合手册打底）
|
||||
|
||||
```
|
||||
专业任务执行时按需叠加通用技能（顺序可调）：
|
||||
|
||||
规划层 plan / persistent-planning / writing-plans / pre-mortem / kanban-orchestrator / team-composer
|
||||
执行层 专业技能（该任务的领域技能，如 KOCR/织忆/ComfyUI/公文…）
|
||||
质量层 三思/行动前验证 → systematic-debugging / requesting-code-review / daemon-code-quality
|
||||
质疑层 反问/批判 → strategy-red-team / pre-mortem / kanban-review-gate / agent-feedback-control
|
||||
验收层 验收证据 → kanban-review-gate / sdlc-review / skill-curator-workflow
|
||||
沉淀层 复盘/技能化 → retro / hermes-self-improvement / daily-recap
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 3. 分类统计（health 按目录 87 类，top 25）
|
||||
|
||||
| 分类 | 技能数 | 平均分 |
|
||||
|---|---|---|
|
||||
| product-management | 48 | 3.6 |
|
||||
| devops | 28 | 5.2 |
|
||||
| creative | 23 | 5.7 |
|
||||
| productivity | 20 | 5.2 |
|
||||
| mlops | 18 | 5.5 |
|
||||
| software-development | 17 | 5.6 |
|
||||
| projectManagerBreakout | 15 | 4.1 |
|
||||
| research | 13 | 5.5 |
|
||||
| autonomous-ai-agents | 11 | 5.5 |
|
||||
| media | 6 | 5.8 |
|
||||
| github | 6 | 5.8 |
|
||||
| knowledge | 5 | 4.8 |
|
||||
| zine | 4 | 3.9 |
|
||||
| social-media | 4 | 5.8 |
|
||||
| automation | 3 | 6.0 |
|
||||
| dashi-ppt | 3 | 4.1 |
|
||||
| openclaw | 3 | 6.4 |
|
||||
| engineering | 2 | 4.5 |
|
||||
| email | 2 | 5.7 |
|
||||
| zhiyi | 2 | 6.3 |
|
||||
| agent-native | 2 | 6.3 |
|
||||
| content | 2 | 2.9 |
|
||||
| product-research | 2 | 5.4 |
|
||||
| community-ops | 1 | 6.0 |
|
||||
| photo-abstract-editorial | 1 | 3.9 |
|
||||
|
||||
> 完整 87 类见 skill-health.json → categories。curator 的 frontmatter category 只有 16 个非空（其余 257 uncategorized），说明 **frontmatter category 大量缺失**（见 §5.2）。
|
||||
|
||||
---
|
||||
|
||||
## 4. D 级 + Tier3 归档候选清单（逐项证据，零动作）
|
||||
|
||||
> 判定规则（按 skill-curator-workflow + 牧尘 2026-08-30 确认）：
|
||||
> 1. durable 引用（SOUL/AGENTS/cron/外 skill/scripts/其他 profile config）> 0 → **保留（KEEP-REF）**
|
||||
> 2. 属于 pm-skills 家族 / 08-30 恢复集 / zine&creative 导入 / product-management 目录 → **保留（KEEP-CONFIRMED，牧尘已批）**
|
||||
> 3. devops/engineering/运维/应急 → **保留（KEEP-OPS，低频是运维正常属性）**
|
||||
> 4. 其余零引用 → **归档候选（ARCHIVE-CANDIDATE）**
|
||||
|
||||
### 4.1 curator Tier3 候选（32 个）逐项判定
|
||||
|
||||
| 技能 | version | grade | curator score | durable 引用 | 判定 | 依据 |
|
||||
|---|---|---|---|---|---|---|
|
||||
| playwright-trace | - | - | 15 | 0 | 归档候选 | - |
|
||||
| ab-test-analysis | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| brainstorm-okrs | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| customer-journey-map | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| job-stories | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| market-segments | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| north-star-metric | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| positioning-ideas | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| pricing-strategy | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| sentiment-analysis | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| summarize-meeting | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| test-scenarios | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| user-personas | 1.0.0 | C | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| user-segmentation | 1.0.0 | D | 15 | 0 | 保留·牧尘已确认 | - |
|
||||
| file-resource-management | 0.1 | D | 5 | 0 | 保留·运维属性 | - |
|
||||
| go-backend-perf-storm-debugging | 0.1 | C | 5 | 0 | 保留·运维属性 | - |
|
||||
| model-health-probe | 0.1 | D | 15 | 0 | 保留·运维属性 | - |
|
||||
| competitor-analysis | 1.0.0 | D | 15 | 5 | 保留·有引用 | profiles/research/config.yaml; skills/autonomous-ai-agents/ao-orchestrator/refer |
|
||||
| frontend-slides | 0.1 | D | 15 | 5 | 保留·有引用 | skills/autonomous-ai-agents/hermes-self-improvement/SKILL.md; skills/curator-fix |
|
||||
| gtm-strategy | 1.0.0 | D | 15 | 3 | 保留·有引用 | skills/knowledge/project-toolbox/SKILL.md; skills/knowledge/skill-library-portin |
|
||||
| guizang-ppt-skill | 0.1 | C | 15 | 5 | 保留·有引用 | skills/autonomous-ai-agents/hermes-self-improvement/SKILL.md; skills/curator-fix |
|
||||
| kanban-review-gate | 1.0 | D | 15 | 2 | 保留·有引用 | skills/devops/self-healing-infrastructure/SKILL.md; skills/devops/self-healing-i |
|
||||
| lean-canvas | 1.0.0 | D | 15 | 3 | 保留·有引用 | scripts/skill-curator.py; skills/knowledge/project-toolbox/SKILL.md |
|
||||
| marketing-ideas | 1.0.0 | D | 15 | 1 | 保留·有引用 | skills/marketing-disciplines/SKILL.md |
|
||||
| pestle-analysis | 1.0.0 | D | 15 | 1 | 保留·有引用 | skills/knowledge/skill-library-porting/SKILL.md |
|
||||
| product-name | 1.0.0 | D | 15 | 2 | 保留·有引用 | skills/product-management/create-prd/SKILL.md; skills/product-management/pre-mor |
|
||||
| product-vision | 1.0.0 | D | 15 | 3 | 保留·有引用 | skills/brainstorm-okrs/SKILL.md; skills/product-management/outcome-roadmap/SKILL |
|
||||
| project-discovery | - | - | 15 | 9 | 保留·有引用 | skills/autonomous-ai-agents/hermes-self-improvement/SKILL.md; skills/curator-fix |
|
||||
| sprint-plan | 1.0.0 | D | 15 | 3 | 保留·有引用 | skills/devops/self-healing-infrastructure/references/skill-curator-false-overlap |
|
||||
| sql-queries | 1.0.0 | D | 15 | 1 | 保留·有引用 | skills/knowledge/skill-library-porting/SKILL.md |
|
||||
| swot-analysis | 1.0.0 | C | 15 | 1 | 保留·有引用 | skills/knowledge/skill-library-porting/SKILL.md |
|
||||
| user-stories | 1.0.0 | D | 15 | 2 | 保留·有引用 | skills/product-management/wwas/SKILL.md; skills/skill-curator-workflow/SKILL.md |
|
||||
|
||||
**注（非 KEEP-REF 的 8 个说明）：**
|
||||
|
||||
1. `playwright-trace` — **幽灵名**。磁盘 skills/ 下无此目录（search 仅命中 stock-research/browser-automation 的 reference 文件）。curator 因它藏在某处 frontmatter 被扫到，health 无此 active 记录 → 建议从 curator 忽略清单 + 确认源后清理扫描入口，**不涉及真实技能**。
|
||||
2. `project-discovery`（curator 名） = `github-project-discovery`（health 名，frontmatter name=project-discovery）→ 是同一技能，**改名对齐**问题不是归档问题。有 9 处 durable 引用（hermes-self-improvement/curator-fixes/self-healing 等）→ 保留。
|
||||
3. `file-resource-management`、`go-backend-perf-storm-debugging`、`model-health-probe`、`cron-watchdog-hygiene`(D) 等 → 运维/应急技能，低频是正常属性；`model-health-probe` 对应 cron「模型健康巡检」(model-health.py)，保留。
|
||||
4. pm-skills 家族（job-stories/market-segments/user-segmentation/brainstorm-okrs/ab-test-analysis/test-scenarios 等，score 15 + version 1.0.0）→ 牧尘 2026-08-30 确认「以后会用」保留（phuryn/pm-skills 25K★ 导入集），curator 的 tier3 是**按 score 机械分档，不是删除信号**。
|
||||
5. `frontend-slides`/`guizang-ppt-skill`/`photo-abstract-editorial` → 2026-08-30 误删后 git 恢复，牧尘确认保留（hermes-self-improvement/curator-fixes 记录）。
|
||||
|
||||
### 4.2 归档候选最终结论（D∪T3 114 个真实技能）
|
||||
|
||||
| 判定 | 数量 | 含义 |
|
||||
|---|---|---|
|
||||
| KEEP-REF | 55 | 有 durable 引用，保留 |
|
||||
| KEEP-CONFIRMED | 47 | pm-skills/恢复集/creative 导入，牧尘已批保留 |
|
||||
| KEEP-OPS | 11 | 运维/应急/工程，低频正常保留 |
|
||||
| ARCHIVE-CANDIDATE | 1 | 零引用无强理由（实为幽灵名 playwright-trace） |
|
||||
|
||||
**一句话结论：D 级 110 与 Tier3 32 的并集（114 真实技能）里，没有需要真归档的对象** —— 每个都有保留理由（引用 / 运维属性 / 牧尘已确认家族）。curator 把它们标 tier3/D 是**分数机械分档**（小体积+无 refs 文件扣分），不等于「无用」。
|
||||
|
||||
真正该处理的是三类**元数据/工程问题**（不是删除）：
|
||||
|
||||
| # | 问题 | 数量 | 建议动作 |
|
||||
|---|---|---|---|
|
||||
| 1 | node_modules 假技能被扫进 health | 2（skill/trace） | health 扫描排除 dashi-ppt/**/node_modules |
|
||||
| 2 | frontmatter name 与目录名错位 | ≥8 | 修 frontmatter name 对齐目录 |
|
||||
| 3 | 重复技能目录（blocked-page-recovery/plan 双份） | 3 对 | 合并，保留引用多的一份 |
|
||||
|
||||
---
|
||||
|
||||
## 5. 缺元数据清单
|
||||
|
||||
### 5.1 缺 version（磁盘级 36 个）
|
||||
|
||||
- agent-search-apis
|
||||
- chinese-platform-extraction
|
||||
- creative/apple-liquid-glass
|
||||
- creative/baoyu-infographic
|
||||
- creative/creative-ideation (fm=ideation)
|
||||
- cron-ops
|
||||
- dashi-ppt
|
||||
- design-system-discipline
|
||||
- devops/cron-watchdog-hygiene
|
||||
- devops/file-resource-management
|
||||
- devops/go-backend-perf-storm-debugging
|
||||
- devops/hermes-profile-gateway-troubleshooting
|
||||
- devops/llm-gateway-ops
|
||||
- devops/provider-tiering (fm=None)
|
||||
- devops/rsshub-ops
|
||||
- devops/self-healing-infrastructure (fm=None)
|
||||
- devops/sqlite-db-corruption-recovery
|
||||
- frontend-slides
|
||||
- github-project-discovery (fm=project-discovery)
|
||||
- guizang-material-illustration
|
||||
- guizang-ppt-skill
|
||||
- karpathy-code-discipline
|
||||
- marketing-disciplines
|
||||
- mlops/training/axolotl
|
||||
- model-health-probe
|
||||
- nuyoah-image-reverse-prompt
|
||||
- persistent-planning
|
||||
- photo-abstract-editorial
|
||||
- plan
|
||||
- ppt-master
|
||||
- productivity/powerpoint
|
||||
- project-manager-breakout
|
||||
- zine/gc-minimal-zine-poster (fm=...-v0-3)
|
||||
- zine/make-photo-stamp-archive
|
||||
- zine/scene-distillation-zine-v1-3
|
||||
- zine/scenes-gathered-zine-v1-3
|
||||
|
||||
> 注意：缺 version ≠ 应归档。多数是外部导入（zine/creative/mlops/agent-skills 库）或运维脚本型技能。建议：**给常用/被引用的补 version**；零引用且不再维护的（如已归档过的）才考虑归档。
|
||||
|
||||
**修正标注（2026-09-06 run 183 二次核对，与 curator 口径对齐）：**
|
||||
- 上表 36 个 = **top-level `version:` 缺失**（curator 读 frontmatter 顶层的口径）。
|
||||
- 其中 12 个实际有**嵌套 `metadata.version`**（plan / agent-search-apis / design-system-discipline / karpathy-code-discipline / persistent-planning / ppt-master / project-manager-breakout / marketing-disciplines / cron-ops / chinese-platform-extraction / devops/rsshub-ops / nuyoah-image-reverse-prompt 等，version 写在 `metadata:` 子块）→ 不是真缺，只是**位置不合规**，建议晋升到顶层（一行 patch），低风险。
|
||||
- 真正**任何层级都无 version** 的 = 24 个：apple-liquid-glass / baoyu-infographic / creative-ideation / dashi-ppt / cron-watchdog-hygiene / file-resource-management / go-backend-perf-storm-debugging / hermes-profile-gateway-troubleshooting / llm-gateway-ops / provider-tiering / self-healing-infrastructure / sqlite-db-corruption-recovery / frontend-slides / github-project-discovery / guizang-material-illustration / guizang-ppt-skill / axolotl / model-health-probe / photo-abstract-editorial / powerpoint / gc-minimal-zine-poster / make-photo-stamp-archive / scene-distillation-zine-v1-3 / scenes-gathered-zine-v1-3。
|
||||
- curator 自己的 `stale_no_version` 只报了 15 个（zine 4 个、axolotl、dashi-ppt、playwright-cli、sqlite-db-corruption-recovery、cron-watchdog-hygiene、ideation、apple-liquid-glass、baoyu-infographic、powerpoint、photo-abstract-editorial、guizang-material-illustration、make-photo-stamp-archive）→ 是 curator 判定口径（叠加其他条件），非全量。
|
||||
|
||||
### 5.2 缺 frontmatter category
|
||||
|
||||
- health 按目录分类有 87 类（无需 frontmatter）；curator 只读 frontmatter → 290/303 缺 category（257 uncategorized + 33 有值但仅 16 类非空）。
|
||||
- 影响：curator 的 by_category 统计失真。**建议**：skill-manager.py fix 批量把 frontmatter category 补为目录名，或 curator 改用目录名作 category 回退（改 curator 比改 290 个 frontmatter 更省）。
|
||||
|
||||
---
|
||||
|
||||
## 6. 真实重叠对（排除假阳性后）
|
||||
|
||||
> curator 报 20 对（score>25%）。逐对看 common_terms 后：**无「应合并删除」级真重叠**。理由：
|
||||
|
||||
| A | B | score | common_terms | 判定 | 说明 |
|
||||
|---|---|---|---|---|---|
|
||||
| sprint-planning | sprint-plan | 100% | planning,sprint | 真同族，可考虑互相引用不合并 | 见下 |
|
||||
| deepseek-harness-research | dsh-ops | 75% | deepseek,dsh,harness | 真同族，可考虑互相引用不合并 | 见下 |
|
||||
| identify-assumptions-new | identify-assumptions-existing | 75% | identify,product,assumptions | 真同族，可考虑互相引用不合并 | 见下 |
|
||||
| brainstorm-ideas-existing | brainstorm-ideas-new | 75% | product,brainstorm,ideas | 真同族，可考虑互相引用不合并 | 见下 |
|
||||
| idea-to-implementation-doc | idea-to-design-doc | 60% | workflow,doc,idea | 真同族，可考虑互相引用不合并 | 见下 |
|
||||
| manim-video | ascii-video | 60% | video,pipeline,production | 真同族，可考虑互相引用不合并 | 见下 |
|
||||
| cron-ops | cron-watchdog-hygiene | 50% | cron | 弱相关/不同用途 | 见下 |
|
||||
| interview-script | summarize-interview | 50% | interview,customer | 弱相关/不同用途 | 见下 |
|
||||
| brainstorm-experiments-new | brainstorm-experiments-existing | 50% | design,product,experiments | 弱相关/不同用途 | 见下 |
|
||||
| api-design | responsive-design | 50% | design | 弱相关/不同用途 | 见下 |
|
||||
| curator-fixes-2026-08-30 | skill-curator-workflow | 50% | curator | 弱相关/不同用途 | 见下 |
|
||||
| hermes-desktop-kanban | kanban-router | 50% | kanban | 弱相关/不同用途 | 见下 |
|
||||
| agnes-ai | ai-media-generation | 50% | agnes | 弱相关/不同用途 | 见下 |
|
||||
| meeting-action-items | document-to-action-items | 50% | action,items | 弱相关/不同用途 | 见下 |
|
||||
| scenes-gathered-zine-v1-3 | scene-distillation-zine-v1-3 | 43% | zine,v1,3 | 弱相关/不同用途 | 见下 |
|
||||
| local-gpu-inference | comfyui-rtx3050 | 40% | 3050,rtx | 弱相关/不同用途 | 见下 |
|
||||
| soulful-framework | daemon-code-quality | 33% | soulful | 弱相关/不同用途 | 见下 |
|
||||
| idea-to-ui-design-brief | idea-to-design-doc | 33% | design,idea | 弱相关/不同用途 | 见下 |
|
||||
| job-stories | user-stories | 33% | stories | 弱相关/不同用途 | 见下 |
|
||||
| competitor-news-monitor | competitor-analysis | 33% | competitor | 弱相关/不同用途 | 见下 |
|
||||
|
||||
**分组说明：**
|
||||
|
||||
1. **同族但互补，保留双方**：`sprint-planning`(projectManagerBreakout) ↔ `sprint-plan`(外部 sprint 框架) — 词 2 个但属不同来源；`identify-assumptions-new/existing`、`brainstorm-ideas-new/existing`、`brainstorm-experiments-new/existing` — pm-skills 刻意分 new/existing 两场景，保留。
|
||||
2. **同族建议链式引用**：`idea-to-implementation-doc`/`idea-to-design-doc`/`idea-to-ui-design-brief` + `idea-superpowers-suite`（superpowers 系列同一方法论不同产物）→ 已在 idea-superpowers-suite 索引，不必合并。
|
||||
3. **名字像但用途不同（假阳性）**：`cron-ops`↔`cron-watchdog-hygiene`（一个通用 cron 运维、一个事故后审计，共同词仅 cron）；`api-design`↔`responsive-design`（后端 API vs 前端 UI，仅 design 撞词）；`job-stories`↔`user-stories`（不同方法）；`manim-video`↔`ascii-video`（一个数学动画一个终端 ASCII，仅 video 撞词）；`hermes-desktop-kanban`↔`kanban-router`（desktop 构建 vs 路由）；`agnes-ai`↔`ai-media-generation`（底层 API 封装 vs 上层生成工作流，agnes-ai 是 ai-media-generation 的实现依赖，不是重复）。
|
||||
4. `competitor-news-monitor`↔`competitor-analysis` — monitor 是持续监控、analysis 是单次分析框架，互补（后者在 report 中已 KEEP-REF）。
|
||||
5. `deepseek-harness-research`↔`dsh-ops` — research 是调研记录、ops 是运维手册，不同生命周期，保留。
|
||||
|
||||
> 结论：**20 对里没有一对该「删除合并」**；最多是「在 SKILL.md 里互相补 related_skills 链接」。已排除假阳性规则（共同词 <3 或全通用词）。
|
||||
|
||||
---
|
||||
|
||||
## 7. 其他命名/工程问题
|
||||
|
||||
| 问题 | 详情 | 建议 |
|
||||
|---|---|---|
|
||||
| 目录重复 | `web/blocked-page-recovery` + `research/blocked-page-recovery`；`software-development/plan` + 顶层 `plan/` | 查引用后合并到主要目录，保留高引用一份 |
|
||||
| fm name=openclaw bug | `openclaw/openclaw-feishu-debug/SKILL.md` frontmatter name=openclaw（与 openclaw/openclaw 撞名） | 修 frontmatter name → openclaw-feishu-debug |
|
||||
| fm name 错位 | creative-ideation(fm=ideation) / github-project-discovery(fm=project-discovery) / gc-minimal-zine-poster(fm=gc-minimal-zine-poster-v0-3) | 修 frontmatter name 与目录一致 |
|
||||
| mlops 双名 | curator=evaluating-llms-harness/fine-tuning-with-trl/serving-llms-vllm vs health=lm-evaluation-harness/trl-fine-tuning/vllm | fm name 对齐 |
|
||||
| 顶层裸目录 | 57 个 bare top-level skill（health 的 category 就是它自己名字） | 可选：归入 category 子目录；低优先 |
|
||||
|
||||
---
|
||||
|
||||
## 8. 给牧尘的决策请求（等待批准，不自动执行）
|
||||
|
||||
1. **归档候选 0 个真实技能** — 唯一 ARCHIVE-CANDIDATE 是幽灵名 playwright-trace（磁盘不存在）。请确认从 curator 关注清单移除。
|
||||
2. **是否批量补 version**（36 个缺 version，多数是导入库）？建议只补被引用/常用的，避免制造「假维护」。
|
||||
3. **frontmatter name 错位修复**（8 处）— 建议修，低风险。
|
||||
4. **node_modules 假技能排除** — 建议 health 扫描排除。
|
||||
5. **目录重复合并**（blocked-page-recovery/plan）— 需要你点头再动。
|
||||
|
||||
> 所有动作均未执行。本报告只读盘点。
|
||||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: plan
|
||||
version: "0.2.0"
|
||||
description: "Generate a minimal LaunchDarkly SDK integration plan from detected stack: choose SDK type(s), dual-SDK server+client when required, files to change, env conventions. Nested under sdk-install; follows detect, precedes apply."
|
||||
license: Apache-2.0
|
||||
compatibility: Requires completed or equivalent detect context (see sibling detect skill)
|
||||
File diff suppressed because one or more lines are too long
|
|
@ -4,13 +4,15 @@
|
|||
"created_at": "2026-06-21T10:11:56.827964+00:00",
|
||||
"created_by": "agent",
|
||||
"last_patched_at": "2026-06-29T14:33:54.209123+00:00",
|
||||
"last_used_at": "2026-07-14T08:33:25.663215+00:00",
|
||||
"last_viewed_at": "2026-07-14T08:33:25.659718+00:00",
|
||||
"last_reused_patch_generation": 0,
|
||||
"last_used_at": "2026-09-06T02:04:43.787102+00:00",
|
||||
"last_viewed_at": "2026-09-06T02:04:43.778647+00:00",
|
||||
"patch_count": 473,
|
||||
"patch_generation": 0,
|
||||
"pinned": false,
|
||||
"state": "stale",
|
||||
"use_count": 235,
|
||||
"view_count": 235
|
||||
"use_count": 236,
|
||||
"view_count": 236
|
||||
},
|
||||
"agent-cli-builder": {
|
||||
"archived_at": null,
|
||||
|
|
@ -693,31 +695,31 @@
|
|||
"archived_at": null,
|
||||
"created_at": "2026-08-25T13:35:46.395378+00:00",
|
||||
"created_by": "agent",
|
||||
"last_patched_at": "2026-09-05T08:02:14.770823+00:00",
|
||||
"last_reused_patch_generation": 3,
|
||||
"last_used_at": "2026-09-05T08:01:56.123808+00:00",
|
||||
"last_viewed_at": "2026-09-05T08:01:56.107217+00:00",
|
||||
"patch_count": 4,
|
||||
"patch_generation": 4,
|
||||
"last_patched_at": "2026-09-06T00:29:10.772455+00:00",
|
||||
"last_reused_patch_generation": 5,
|
||||
"last_used_at": "2026-09-06T02:21:30.471529+00:00",
|
||||
"last_viewed_at": "2026-09-06T02:21:30.463028+00:00",
|
||||
"patch_count": 5,
|
||||
"patch_generation": 5,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 3,
|
||||
"view_count": 3
|
||||
"use_count": 5,
|
||||
"view_count": 5
|
||||
},
|
||||
"cron-watchdog-hygiene": {
|
||||
"archived_at": null,
|
||||
"created_at": "2026-09-03T15:13:36.558361+00:00",
|
||||
"created_by": "agent",
|
||||
"last_patched_at": "2026-09-05T13:44:51.703100+00:00",
|
||||
"last_reused_patch_generation": 11,
|
||||
"last_used_at": "2026-09-05T13:43:57.008101+00:00",
|
||||
"last_viewed_at": "2026-09-05T13:43:56.997668+00:00",
|
||||
"patch_count": 12,
|
||||
"patch_generation": 12,
|
||||
"last_patched_at": "2026-09-06T02:16:17.413343+00:00",
|
||||
"last_reused_patch_generation": 13,
|
||||
"last_used_at": "2026-09-06T02:16:09.372107+00:00",
|
||||
"last_viewed_at": "2026-09-06T02:16:09.312477+00:00",
|
||||
"patch_count": 15,
|
||||
"patch_generation": 15,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 9,
|
||||
"view_count": 9
|
||||
"use_count": 11,
|
||||
"view_count": 11
|
||||
},
|
||||
"curator-fixes-2026-08-30": {
|
||||
"archived_at": null,
|
||||
|
|
@ -740,14 +742,14 @@
|
|||
"created_by": "agent",
|
||||
"last_patched_at": "2026-08-19T15:34:45.023034+00:00",
|
||||
"last_reused_patch_generation": 3,
|
||||
"last_used_at": "2026-09-05T16:38:23.973164+00:00",
|
||||
"last_viewed_at": "2026-09-05T16:38:23.960631+00:00",
|
||||
"last_used_at": "2026-09-06T02:24:42.307574+00:00",
|
||||
"last_viewed_at": "2026-09-06T02:24:42.299632+00:00",
|
||||
"patch_count": 43,
|
||||
"patch_generation": 3,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 30,
|
||||
"view_count": 30
|
||||
"use_count": 31,
|
||||
"view_count": 31
|
||||
},
|
||||
"daily-recap": {
|
||||
"archived_at": null,
|
||||
|
|
@ -1618,14 +1620,14 @@
|
|||
"created_by": null,
|
||||
"last_patched_at": "2026-08-28T13:02:49.751591+00:00",
|
||||
"last_reused_patch_generation": 2,
|
||||
"last_used_at": "2026-09-04T17:27:21.253941+00:00",
|
||||
"last_viewed_at": "2026-09-04T17:27:21.241758+00:00",
|
||||
"last_used_at": "2026-09-06T02:56:48.269666+00:00",
|
||||
"last_viewed_at": "2026-09-06T02:56:48.264889+00:00",
|
||||
"patch_count": 125,
|
||||
"patch_generation": 2,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 167,
|
||||
"view_count": 166
|
||||
"use_count": 168,
|
||||
"view_count": 167
|
||||
},
|
||||
"hermes-desktop-kanban": {
|
||||
"archived_at": null,
|
||||
|
|
@ -1702,14 +1704,14 @@
|
|||
"created_by": null,
|
||||
"last_patched_at": "2026-08-12T05:13:32.138839+00:00",
|
||||
"last_reused_patch_generation": 1,
|
||||
"last_used_at": "2026-09-03T13:43:48.014380+00:00",
|
||||
"last_viewed_at": "2026-09-03T13:43:48.001778+00:00",
|
||||
"last_used_at": "2026-09-06T03:25:42.511600+00:00",
|
||||
"last_viewed_at": "2026-09-06T03:25:42.502653+00:00",
|
||||
"patch_count": 81,
|
||||
"patch_generation": 1,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 144,
|
||||
"view_count": 133
|
||||
"use_count": 152,
|
||||
"view_count": 137
|
||||
},
|
||||
"hermes-venv-dependency-safety": {
|
||||
"archived_at": null,
|
||||
|
|
@ -1851,14 +1853,14 @@
|
|||
"created_by": null,
|
||||
"last_patched_at": null,
|
||||
"last_reused_patch_generation": 0,
|
||||
"last_used_at": "2026-09-05T16:45:03.386446+00:00",
|
||||
"last_viewed_at": "2026-09-05T16:45:03.374054+00:00",
|
||||
"last_used_at": "2026-09-06T03:24:46.295113+00:00",
|
||||
"last_viewed_at": "2026-09-06T03:24:46.278351+00:00",
|
||||
"patch_count": 0,
|
||||
"patch_generation": 0,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 19,
|
||||
"view_count": 19
|
||||
"use_count": 21,
|
||||
"view_count": 21
|
||||
},
|
||||
"kanban-review-gate": {
|
||||
"archived_at": null,
|
||||
|
|
@ -1866,25 +1868,25 @@
|
|||
"created_by": null,
|
||||
"last_patched_at": null,
|
||||
"last_reused_patch_generation": 0,
|
||||
"last_used_at": "2026-09-05T17:11:15.745837+00:00",
|
||||
"last_viewed_at": "2026-09-05T17:11:15.740990+00:00",
|
||||
"last_used_at": "2026-09-06T03:24:46.299752+00:00",
|
||||
"last_viewed_at": "2026-09-06T03:24:46.290519+00:00",
|
||||
"patch_count": 0,
|
||||
"patch_generation": 0,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 3,
|
||||
"view_count": 3
|
||||
"use_count": 4,
|
||||
"view_count": 4
|
||||
},
|
||||
"kanban-router": {
|
||||
"archived_at": null,
|
||||
"created_at": "2026-09-01T15:56:37.956717+00:00",
|
||||
"created_by": null,
|
||||
"last_patched_at": null,
|
||||
"last_patched_at": "2026-09-06T02:06:02.180016+00:00",
|
||||
"last_reused_patch_generation": 0,
|
||||
"last_used_at": "2026-09-05T16:38:14.991475+00:00",
|
||||
"last_viewed_at": "2026-09-05T16:38:14.986646+00:00",
|
||||
"patch_count": 0,
|
||||
"patch_generation": 0,
|
||||
"patch_count": 2,
|
||||
"patch_generation": 2,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 6,
|
||||
|
|
@ -1894,16 +1896,16 @@
|
|||
"archived_at": null,
|
||||
"created_at": "2026-09-01T13:21:22.328206+00:00",
|
||||
"created_by": "agent",
|
||||
"last_patched_at": "2026-09-02T04:45:15.989453+00:00",
|
||||
"last_reused_patch_generation": 3,
|
||||
"last_used_at": "2026-09-02T04:44:53.449322+00:00",
|
||||
"last_viewed_at": "2026-09-02T04:44:53.444331+00:00",
|
||||
"patch_count": 4,
|
||||
"patch_generation": 4,
|
||||
"last_patched_at": "2026-09-06T02:23:13.775540+00:00",
|
||||
"last_reused_patch_generation": 4,
|
||||
"last_used_at": "2026-09-06T02:23:04.572011+00:00",
|
||||
"last_viewed_at": "2026-09-06T02:23:04.467821+00:00",
|
||||
"patch_count": 6,
|
||||
"patch_generation": 6,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 4,
|
||||
"view_count": 4
|
||||
"use_count": 6,
|
||||
"view_count": 6
|
||||
},
|
||||
"kanban-worker": {
|
||||
"archived_at": null,
|
||||
|
|
@ -1911,14 +1913,14 @@
|
|||
"created_by": null,
|
||||
"last_patched_at": null,
|
||||
"last_reused_patch_generation": 0,
|
||||
"last_used_at": "2026-09-05T12:32:20.043040+00:00",
|
||||
"last_viewed_at": "2026-09-05T12:32:20.037453+00:00",
|
||||
"last_used_at": "2026-09-06T03:24:42.216024+00:00",
|
||||
"last_viewed_at": "2026-09-06T03:24:42.207150+00:00",
|
||||
"patch_count": 0,
|
||||
"patch_generation": 0,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 12,
|
||||
"view_count": 12
|
||||
"use_count": 14,
|
||||
"view_count": 14
|
||||
},
|
||||
"karpathy-code-discipline": {
|
||||
"archived_at": null,
|
||||
|
|
@ -1954,16 +1956,16 @@
|
|||
"archived_at": null,
|
||||
"created_at": "2026-08-20T19:37:37.464157+00:00",
|
||||
"created_by": "agent",
|
||||
"last_patched_at": "2026-09-05T10:36:03.116505+00:00",
|
||||
"last_reused_patch_generation": 16,
|
||||
"last_used_at": "2026-09-05T10:35:56.474974+00:00",
|
||||
"last_viewed_at": "2026-09-05T10:35:56.470221+00:00",
|
||||
"patch_count": 17,
|
||||
"patch_generation": 17,
|
||||
"last_patched_at": "2026-09-06T01:57:05.268712+00:00",
|
||||
"last_reused_patch_generation": 17,
|
||||
"last_used_at": "2026-09-06T01:56:53.716632+00:00",
|
||||
"last_viewed_at": "2026-09-06T01:56:53.708928+00:00",
|
||||
"patch_count": 18,
|
||||
"patch_generation": 18,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 26,
|
||||
"view_count": 26
|
||||
"use_count": 27,
|
||||
"view_count": 27
|
||||
},
|
||||
"lazy-senior-dev": {
|
||||
"archived_at": null,
|
||||
|
|
@ -2197,14 +2199,14 @@
|
|||
"created_by": "agent",
|
||||
"last_patched_at": "2026-09-05T10:36:30.971834+00:00",
|
||||
"last_reused_patch_generation": 8,
|
||||
"last_used_at": "2026-09-05T13:11:38.546576+00:00",
|
||||
"last_viewed_at": "2026-09-05T13:11:38.534422+00:00",
|
||||
"last_used_at": "2026-09-06T01:54:37.941761+00:00",
|
||||
"last_viewed_at": "2026-09-06T01:54:37.937327+00:00",
|
||||
"patch_count": 8,
|
||||
"patch_generation": 8,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 9,
|
||||
"view_count": 9
|
||||
"use_count": 10,
|
||||
"view_count": 10
|
||||
},
|
||||
"memory-routing-contract": {
|
||||
"archived_at": null,
|
||||
|
|
@ -3084,14 +3086,14 @@
|
|||
"created_by": null,
|
||||
"last_patched_at": null,
|
||||
"last_reused_patch_generation": 0,
|
||||
"last_used_at": "2026-08-30T09:03:16.603249+00:00",
|
||||
"last_viewed_at": "2026-08-30T09:03:16.588728+00:00",
|
||||
"last_used_at": "2026-09-06T03:25:42.506977+00:00",
|
||||
"last_viewed_at": "2026-09-06T03:25:42.497703+00:00",
|
||||
"patch_count": 0,
|
||||
"patch_generation": 0,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 3,
|
||||
"view_count": 3
|
||||
"use_count": 10,
|
||||
"view_count": 7
|
||||
},
|
||||
"skill-library-porting": {
|
||||
"archived_at": null,
|
||||
|
|
@ -3209,14 +3211,14 @@
|
|||
"created_by": null,
|
||||
"last_patched_at": "2026-09-05T07:33:21.286986+00:00",
|
||||
"last_reused_patch_generation": 3,
|
||||
"last_used_at": "2026-09-05T12:32:10.767063+00:00",
|
||||
"last_viewed_at": "2026-09-05T12:32:10.762235+00:00",
|
||||
"last_used_at": "2026-09-06T01:26:16.489438+00:00",
|
||||
"last_viewed_at": "2026-09-06T01:26:16.483508+00:00",
|
||||
"patch_count": 3,
|
||||
"patch_generation": 3,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 53,
|
||||
"view_count": 53
|
||||
"use_count": 57,
|
||||
"view_count": 57
|
||||
},
|
||||
"stock-research": {
|
||||
"archived_at": null,
|
||||
|
|
@ -3296,13 +3298,15 @@
|
|||
"created_at": "2026-07-08T18:05:11.777769+00:00",
|
||||
"created_by": null,
|
||||
"last_patched_at": "2026-07-08T18:53:51.572745+00:00",
|
||||
"last_used_at": "2026-07-09T00:07:14.571066+00:00",
|
||||
"last_viewed_at": "2026-07-09T00:07:14.562274+00:00",
|
||||
"last_reused_patch_generation": 0,
|
||||
"last_used_at": "2026-09-06T02:55:24.388345+00:00",
|
||||
"last_viewed_at": "2026-09-06T02:23:21.366427+00:00",
|
||||
"patch_count": 9,
|
||||
"patch_generation": 0,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 8,
|
||||
"view_count": 8
|
||||
"use_count": 13,
|
||||
"view_count": 10
|
||||
},
|
||||
"teams-meeting-pipeline": {
|
||||
"archived_at": null,
|
||||
|
|
@ -3595,16 +3599,16 @@
|
|||
"archived_at": null,
|
||||
"created_at": "2026-05-29T19:39:03.373231+00:00",
|
||||
"created_by": null,
|
||||
"last_patched_at": "2026-09-05T15:11:20.846671+00:00",
|
||||
"last_patched_at": "2026-09-06T01:52:26.120226+00:00",
|
||||
"last_reused_patch_generation": 10,
|
||||
"last_used_at": "2026-09-05T18:36:44.523189+00:00",
|
||||
"last_viewed_at": "2026-09-05T18:36:44.518207+00:00",
|
||||
"patch_count": 741,
|
||||
"patch_generation": 10,
|
||||
"last_used_at": "2026-09-06T01:15:06.519755+00:00",
|
||||
"last_viewed_at": "2026-09-06T01:15:06.509545+00:00",
|
||||
"patch_count": 743,
|
||||
"patch_generation": 12,
|
||||
"pinned": false,
|
||||
"state": "active",
|
||||
"use_count": 461,
|
||||
"view_count": 435
|
||||
"use_count": 464,
|
||||
"view_count": 438
|
||||
},
|
||||
"zhiyi-dev": {
|
||||
"archived_at": null,
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: agent-search-apis
|
||||
version: "1.0.0"
|
||||
description: "Use when searching for code, finance, legal, or vertical."
|
||||
metadata:
|
||||
version: "1.0.0"
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: chinese-platform-extraction
|
||||
version: "1.0.0"
|
||||
description: "Use when extracting content from WeChat or Zhihu."
|
||||
metadata:
|
||||
version: "1.0.0"
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: apple-liquid-glass
|
||||
version: "1.0.0"
|
||||
description: Build Apple-grade web UI — the "macOS Liquid Glass" aesthetic. Use when designing or building any page, component, or interface that should feel like the latest Apple (apple.com / Apple Newsroom / macOS): light grey-white ground, unified white surfaces with hairline dividers, restrained typography with negative tracking, frosted glass used only where layers overlap, and color reserved for accent. Covers fluid motion for interactive layers (springs, interruptibility, reduced-motion). Portable across projects (any stack) — ships tokens, copy-paste components, line icons, a reference page, and a self-check. Trigger on "Apple style", "liquid glass", "make it look like apple.com", "clean Apple UI", "苹果风", "液态玻璃".
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: baoyu-infographic
|
||||
version: "1.0.0"
|
||||
tags: [baoyu infographic, creative]
|
||||
description: "信息图生成器 — 21种布局(Layout) × 21种风格(Style)自由组合。布局: bento-grid/linear-progression/binary-comparison/hierarchical-layers/tree-branching 等。风格: craft-handmade/claymation/kawaii/cyberpunk/technical-schematic 等。图片来源: image_generate工具。"
|
||||
related_skills:
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: ideation
|
||||
name: creative-ideation
|
||||
version: "1.0.0"
|
||||
description: "Creative Ideation — 约束驱动的项目创意生成器。21种创意约束(自动化的烦人小事/CUI工具/弗兰肯斯坦周/敌意UI/搞笑结局等)，按用户输入匹配最佳约束→生成3个具体idea→选定即开建。"
|
||||
related_skills:
|
||||
- baoyu-infographic: "信息图可视化 — 把创意变成视觉作品"
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: cron-ops
|
||||
version: "1.0.0"
|
||||
description: "Use when managing cron jobs or configuring fallback chains."
|
||||
metadata:
|
||||
version: "1.0.0"
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: dashi-ppt
|
||||
version: "1.0.0"
|
||||
description: 制作 PPT、演示文稿、幻灯片、汇报材料时使用。Dashi PPT 基于预置视觉主题组合页面,生成可离线打开、可在浏览器编辑的 HTML 演示,支持导出 PPTX / PDF 文件。
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: design-system-discipline
|
||||
version: "1.0.0"
|
||||
description: Use when generating UI/UX code, design systems, slides, or banner design. Three-layer token architecture (primitive→semantic→component), professional component specs, design-to-code handoff discipline. Borrowed from ui-ux-pro-max-skill.
|
||||
license: MIT
|
||||
metadata:
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: cron-watchdog-hygiene
|
||||
version: "1.0.0"
|
||||
description: Use when 事故后审计/清理定时任务与看门狗。去重cron、除破坏性自愈(删WAL/kill-9)、清误报源。
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: file-resource-management
|
||||
version: "1.0.0"
|
||||
description: 文件资源管理机制 — 脚本/图片/视频/文档分类存放，避免根目录散落。触发词"文件乱/资源管理/整理文件/归类"。
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: go-backend-perf-storm-debugging
|
||||
version: "1.0.0"
|
||||
description: Use when Go 后端 CPU高/超时/异常。六坑：http无超时/O(V²)/零值触发风暴/超时误报/缓存当数据源/Rust编译并行OOM。
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: hermes-profile-gateway-troubleshooting
|
||||
version: "1.0.0"
|
||||
description: 诊断和修复 Hermes 多 profile 分身 gateway 的问题，包括配置路径、模型切换、重启等
|
||||
category: devops
|
||||
---
|
||||
|
|
|
|||
|
|
@ -232,7 +232,11 @@ ls /var/lib/memoryweave/episodes.lance/_versions/ | wc -l # 18630 版本 / 1.
|
|||
|
||||
**⚠️ 与"禁 Python 写"铁律的关系**：铁律指 Python lancedb **写 records**（encodings21 格式 Rust 0.22 读不了）。Python 侧跑 optimize/VACUUM 实测安全——重写文件后 Rust lance 0.22 sidecar 正常启动、search/写入全通（2026-09-05 版本对验证）。执行时仍停 zhiyid+consolidate + 快照兜底。
|
||||
|
||||
**✅ Python `tbl.delete(where)` 按条件直删也安全（2026-09-05 实测）**：删除 8465 条噪音（content LIKE 三类循环同步状态）后重启 sidecar，stats/recall 全正常——delete 与 optimize 同理（重写数据文件为标准 lance 格式，Rust 兼容）。批量清噪音的标准路径：停 zhiyid+consolidate → 硬链接快照 → `tbl.delete("content LIKE '%模式%' OR ...")` → 重启 → recall 验证。**⚠️ 千万别用 zhiyid 的 `DELETE /api/v1/distilled/{id}` API 清数据**——它是假删除（见 zhiyi skill 2026-09-05 节），只标记 Go 内存 map，重启即失效。
|
||||
**✅ Python `tbl.delete(where)` 按条件直删也安全（2026-09-05 实测）**：删除 8465 条噪音（content LIKE 三类循环同步状态）后重启 sidecar，stats/recall 全正常——delete 与 optimize 同理（重写数据文件为标准 lance 格式，Rust 兼容）。批量清噪音的标准路径：停 zhiyid+consolidate → 硬链接快照 → `tbl.delete("content LIKE '%模式%' OR ...")` → 重启 → recall 验证。
|
||||
|
||||
**⚠️ SoftDelete API 已持久化（2026-09-06 f6cd93f 起）**：`DELETE /api/v1/distilled/{id}` 现在真删 = is_deleted=true 落库（memories 表内标记）+ 重启不复活（实测：SoftDelete persisted 日志 + recall 0）。旧版"假删除只标记 Go 内存"警告已过时——zhiyid 日志 `SoftDelete persisted` 即持久化成功。
|
||||
|
||||
**✅ 软删残留物理清理（2026-09-06 实测）**：SoftDelete 标记的行（is_deleted=true）仍占 memories 表、让 stats total_memories 虚高。物理移除：停 zhiyid+consolidate → 硬链接快照 → `tbl.delete("is_deleted = true")` → 重启 → 验证。实测 3094→2264（830 条软删残留移除），stats 恢复干净语义（memories=全有效、tombstone=0）。**有效记忆 = total_memories − tombstone_count**（清理前 3075−830=2245 才是真实有效数）。
|
||||
|
||||
完整实录见 `references/vacuum-compact-20260905.md`
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: llm-gateway-ops
|
||||
version: "1.0.0"
|
||||
description: LLM 网关运维 — OmniRoute/NewAPI 类 AI 网关的评估、部署、API 调用、观测体系、cron 切换。含网关选型方法论、npm 安装坑、urllib 不兼容、SSE 读取模式、systemd 常驻、看门狗接入。
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: rsshub-ops
|
||||
version: "1.0.0"
|
||||
description: "RSSHub 运维 — 部署、GitHub搜索修复、RSS源管理、织忆对接"
|
||||
metadata:
|
||||
version: "1.0.0"
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: sqlite-db-corruption-recovery
|
||||
version: "1.0.0"
|
||||
description: Use when SQLite/state.db 报损坏/malformed/corruption。11 种损坏模式分类排查 + .recover/VACUUM INTO 重建 + 误判循环规避。
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: frontend-slides
|
||||
version: "1.0.0"
|
||||
description: Create stunning, animation-rich HTML presentations from scratch or by converting PowerPoint files. Use when the user wants to build a presentation, convert a PPT/PPTX to web, or create slides for a talk/pitch. Helps non-designers discover their aesthetic through visual exploration rather than abstract choices.
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -2,7 +2,8 @@
|
|||
created: 2025-12-16
|
||||
modified: 2026-06-05
|
||||
reviewed: 2026-06-05
|
||||
name: project-discovery
|
||||
name: github-project-discovery
|
||||
version: "1.0.0"
|
||||
description: Project orientation for unfamiliar codebases. Use when entering a new project, exploring unknown repos, or working on shaky assumptions about build, test, lint, or CI setup.
|
||||
user-invocable: false
|
||||
allowed-tools: Bash(bash *), Read, Grep, Glob, TodoWrite
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: guizang-material-illustration
|
||||
version: "1.0.0"
|
||||
description: Generate Guizang-style material illustrations, labeled explanatory visuals, material-styled chart illustrations, and data-first editorial images from articles, notes, product concepts, workplace reports, creator posts, tutorials, school materials, humanities topics, science explanations, screenshots, or chart data. Use when the user asks for 配图, 带字插图, 解释图, 图解插画, 概念拆解图, 图表美化, 数据图美化, 3D 图表, 汇报配图, 内容配图, 小学课文配图, 生物/化学/物理解释图, 人文类配图, process/loop/system diagrams, or wants GPT-Image / image generation to create supporting images that can sit inside social cards, docs, slides, PPTs, or posts.
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: guizang-ppt-skill
|
||||
version: "1.0.0"
|
||||
description: 生成横向翻页网页 PPT（单 HTML 文件），含 WebGL 背景、演讲者视图、观众屏同步、讲稿备注、章节幕封、数据大字报、图片网格等模板。提供两种风格：① "电子杂志 × 电子墨水"（衬线 + 流体背景 + 暖色） ② "瑞士国际主义"（无衬线 + 网格点阵 + IKB/柠檬黄/柠檬绿/安全橙高亮）。当用户需要制作分享 / 演讲 / 发布会风格的网页 PPT，或提到"杂志风 PPT"、"瑞士风 PPT"、"Swiss Style"、"horizontal swipe deck"时使用。
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: karpathy-code-discipline
|
||||
version: "1.0.0"
|
||||
description: Use when writing, reviewing, or refactoring code. Karpathy-inspired coding discipline — minimum code, surgical changes, surface tradeoffs, goal-driven verification. 4 条核心原则防止 LLM 编码陷阱。适用所有写代码场景。
|
||||
license: MIT
|
||||
metadata:
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: marketing-disciplines
|
||||
version: "1.0.0"
|
||||
description: 营销/SEO/CRO/文案方法论 + 177 个科研工具索引。Use when handling marketing tasks (CRO/copywriting/SEO/growth) OR scientific/medical computing (bioinformatics/chemistry/genomics). Borrowed from coreyhaines31/marketingskills (33K⭐) + InnoLabs/claude-scientific-skills (9100⭐).
|
||||
license: MIT
|
||||
metadata:
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
---
|
||||
name: evaluating-llms-harness
|
||||
name: lm-evaluation-harness
|
||||
description: "lm-eval-harness: benchmark LLMs (MMLU, GSM8K, etc.)."
|
||||
version: 1.0.0
|
||||
author: Orchestra Research
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
---
|
||||
name: serving-llms-vllm
|
||||
name: vllm
|
||||
description: "vLLM: high-throughput LLM serving, OpenAI API, quantization."
|
||||
version: 1.0.0
|
||||
author: Orchestra Research
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: axolotl
|
||||
version: "1.0.0"
|
||||
tags: [axolotl, mlops]
|
||||
description: "Axolotl: YAML配置驱动的LLM微调框架 — LoRA/QLoRA/DPO/KTO/ORPO/GRPO全支持，100+模型兼容，DeepSpeed/FSDP多卡训练。依赖: axolotl, torch, transformers, datasets, peft, accelerate, deepspeed。"
|
||||
related_skills:
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
---
|
||||
name: fine-tuning-with-trl
|
||||
name: trl-fine-tuning
|
||||
description: "TRL: SFT, DPO, PPO, GRPO, reward modeling for LLM RLHF."
|
||||
version: 1.0.0
|
||||
author: Orchestra Research
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: model-health-probe
|
||||
version: "1.0.0"
|
||||
description: Probe all LLM providers + models in prof-b config for live health (HTTP code + latency). Use when reporting model availability, diagnosing 429/timeout cascades, validating a fallback chain, or before declaring a routing decision safe. Outputs stable one-line-per-probe format suitable for diff-monitoring via cron.
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: nuyoah-image-reverse-prompt
|
||||
version: "0.1.0"
|
||||
description: Use when the user provides a reference image and asks to 图片反推、画面解构、提取结构词、分析构图色彩材质、反推可直接生图的中文提示词，明确要求通过生成返图来校准本 Skill，或明确要求检查、更新“南鸢图片反推 Skill”本身。普通调用只交付结构拆解、参考色卡和可复制 prompt；不用于跳过分析直接生图、普通图片编辑或无参考图的通用提示词写作。
|
||||
license: MIT
|
||||
metadata:
|
||||
|
|
|
|||
|
|
@ -1,6 +1,6 @@
|
|||
---
|
||||
tags: [openclaw]
|
||||
name: openclaw
|
||||
name: openclaw-feishu-debug
|
||||
description: "OpenClaw 日常维护与故障排查 — skill 管理、配置修复、飞书频道、诊断命令"
|
||||
version: 1.1.0
|
||||
author: 小唯
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: persistent-planning
|
||||
version: "1.0.0"
|
||||
description: Manus-style persistent file-based planning — keep task_plan.md, findings.md, progress.md on disk so work survives /clear, compaction, and long sessions. Use when planning multi-step projects, research tasks, or work requiring 5+ tool calls. Supports session recovery.
|
||||
license: MIT
|
||||
metadata:
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: ppt-master
|
||||
version: "4.8.0"
|
||||
description: "Use when creating/editing PPT/PPTX presentations."
|
||||
metadata:
|
||||
version: "4.8.0"
|
||||
|
|
|
|||
|
|
@ -1,6 +1,7 @@
|
|||
---
|
||||
tags: [productivity]
|
||||
name: powerpoint
|
||||
version: "1.0.0"
|
||||
description: "Create, read, edit .pptx decks, slides, notes, templates."
|
||||
license: Proprietary. LICENSE.txt has complete terms
|
||||
platforms: [linux, macos, windows]
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: project-manager-breakout
|
||||
version: "1.0.0"
|
||||
description: "Use when breaking down projects or planning sprints."
|
||||
metadata:
|
||||
version: "1.0.0"
|
||||
|
|
|
|||
|
|
@ -1,3 +0,0 @@
|
|||
---
|
||||
description: Skills for reaching web content when direct access fails — blocked, paywalled, rate-limited, or bot-walled pages.
|
||||
---
|
||||
|
|
@ -2,8 +2,8 @@
|
|||
tags: [zhiyi]
|
||||
name: zhiyi
|
||||
description: "织忆 (MemoryWeave) 聚合技能 — API 客户端 + 开发工作流 + 运维规范。含 commit/recall API、数据架构、部署验证、Go 方法论。"
|
||||
version: 11.39
|
||||
updated: 2026-09-05（memory_feedback 闭环修复：prefetch/memory_search 带 memory_id + system prompt 引导标记）
|
||||
version: 11.41
|
||||
updated: 2026-09-06（tombstone_count 统计源修复 + f6cd93f P2 合流记录）
|
||||
|
||||
---
|
||||
|
||||
|
|
@ -40,7 +40,31 @@ EOF
|
|||
|
||||
---
|
||||
|
||||
## ⚠️ 遗忘机制闭环修复（2026-09-05 深夜，kanban t_883d033a，commit febc2c9）
|
||||
## ⚠️ tombstone_count 统计源修复 + P2 合流（2026-09-06，commit f6cd93f/f6c4383/1bf23c4）
|
||||
|
||||
**问题**：`GET /api/v1/stats` 的 `tombstone_count` 恒 0。根因：SoftDelete 持久化语义 = memories 表内标记 `is_deleted=true`（不写独立 tombstones 表），但 Rust `stats()` 原数的是**空的 tombstones 表** → 恒 0 指标失真。
|
||||
|
||||
**修复**（rust/src/lancedb_ops.rs stats()）：对 memories 表 `count_rows(Some("is_deleted = true".to_string()))` 作为 tombstone_count。
|
||||
|
||||
**🔴 count_rows filter 类型坑**：此 lance 版本 `count_rows` 签名是 `Option<String>` 不是 `Option<&str>`——写 `Some("is_deleted = true")` 编译报 E0308，必须 `.to_string()`。
|
||||
|
||||
**有效记忆计算**：`total_memories - tombstone_count`（2026-09-06 实测：3075 - 830 = **2245 条真实有效**，之前 total 虚高含 830 条软删噪音残留）。
|
||||
|
||||
**P2 合流（remote main 2026-09-06 02:32）**：遗忘机制最终版已在 remote：
|
||||
- `4376f1f` P1 防风暴（限长/分桶/抽样下推/decay 分批）
|
||||
- `e67f0bc` Embedder http client 8s 超时 + decay 分批 AC-1 日志
|
||||
- `6c1c21f` PageRank O(V+E)
|
||||
- `f6cd93f` **P2 候选源升级**：GetCandidatesForForgetting 走 Rust `lancedb_scan`（安全版：limit 2000 + 跳 vector + 真读 last_recalled_at + 消费方截断护栏），替代纯缓存遍历（缓存空 → 0 候选 → 9-25 到期记忆永不忘的 bug）。IPC 失败才 fallback 缓存。
|
||||
|
||||
**部署状态（2026-09-06 09:25）**：`/home/muc/bin/zhiyid-new` = f6cd93f 源码构建；`/home/muc/bin/zhiyi-consolidate` = 1bf23c4（含 tombstone fix）。验证：`[decay] scanned=2000 forgotten=0` + stats tombstone_count=830。
|
||||
|
||||
**🔴 血泪教训（2026-09-06）**：
|
||||
1. **cargo build 输出别用 grep 管道吞**：`cargo build 2>&1 | grep error | tail` 在编译错误时 grep 可能不匹配/被 SIGPIPE，脚本照样 `cp` 旧产物 + echo 成功——掩盖失败（本次 build-rust3 EXIT=101 被吞，部署了旧 binary）。正确：完整日志落文件，检查 `EXIT=$?`。
|
||||
2. **本地 /tmp/memoryweave 常落后 remote**：本次本地 checkout 停在 4c62074，remote 已有 4 个新 commit（P1+P2）——手动恢复 febc2c9 做重复劳动。**动源码前先 `git pull`**。
|
||||
3. **运行中 binary 不能 cp**（Text file busy）：先 `systemctl --user stop`（主动 stop 不触发 Restart=always）→ cp → start。
|
||||
4. **febc2c9 风暴真根因** = server.go consolidate O(n²) 逐对 LLM（4c62074 已修按目标分组），**不是** lancedb_scan 代码本身——0881b9c 误删 scan 修复是过度回退。
|
||||
|
||||
---
|
||||
|
||||
**症状**：auto_forget 0 条（09-01 至今），记忆只增不减。
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: gc-minimal-zine-poster-v0-3
|
||||
name: gc-minimal-zine-poster
|
||||
version: "1.0.0"
|
||||
description: 极简Zine海报：大留白+小拼贴+彩色点。用「极简海报/gc-minimal-zine-poster」触发。 原文：Generate or analyze poetic paper-texture minimal zine posters with large negative space, a small editorial collage or focal element, experimental typography, and one clear color accent. Use when the user gives a theme, sentence, article, mood, object, photo, content brief, reference images, or an image folder and wants a final poster image, a production-ready image prompt, reusable style rules, or varied layouts that stay in one visual family.
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: make-photo-stamp-archive
|
||||
version: "1.0.0"
|
||||
description: 照片图章档案：保留原照片+定制图章。用「图章档案/photo-stamp-archive」触发。 原文：Transform one or more supplied photos into clean direct-splice archival artworks that pair a faithfully preserved photograph with a warm-white paper panel containing a compact hand-pressed custom seal or stamp. Use for requests mentioning 图章、印章、stamp、seal、照片加图章、照片档案、左右拼接、清爽纸张、圆形章、方形章、异形章、建筑轮廓章、主体缩小、移动到角落, or iterative changes to stamp shape, border, scale, position, ink color, caption, paper age, or splice orientation.
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: scene-distillation-zine-v1-3
|
||||
version: "1.0.0"
|
||||
description: "影像蒸馏海报：照片转情绪化抽象纸刊。用「影像蒸馏/scene-distillation-zine/告别与回望」触发。 原文：Transform a user-supplied photo into an expressive minimal zine poster made only from original source-derived illustration, an artistic proposition, emotional tension, visual metaphor, spacious negative space, art-directed high-chroma color, and unconstrained authorial typography. Let wording, language, amount, placement, type voices, scale, direction, legibility, and image interaction follow expression and aesthetic judgment rather than presets. Preserve source orientation by default with a 3:5 portrait output or 5:3 landscape output. Add source-derived distributed supporting accents and a natural isolated-contour option alongside adaptive paper-edge transitions. Support an exact `单色块模式` trigger for one contiguous saturated color field with all remaining forms in neutral ink. Use for authored abstract or editorial reinterpretations that communicate an emotion or idea without embedding, cropping, tracing, or preserving the original photographic material in the final image."
|
||||
---
|
||||
|
||||
|
|
|
|||
|
|
@ -1,5 +1,6 @@
|
|||
---
|
||||
name: scenes-gathered-zine-v1-3
|
||||
version: "1.0.0"
|
||||
description: "拾景纸刊拼贴海报：照片转Zine风格拼贴。用「拾景纸刊/实景拼贴/scenes-gathered-zine」触发。 原文：Transform a user-supplied photo into a vertical 3:5 Gathered Scenes Zine poster that anchors truthful photography inside a spacious source-derived abstract illustration field, aggressively compresses dense foliage and other micro-detail into a few large quiet forms, integrates one high-chroma hue as compositional structure, and preserves a visibly hand-torn fibrous photo-to-paper edge. Use when the user wants a tactile minimal paper collage with simplified illustration, active negative space, restrained English-default, Chinese, or bilingual micro-text, and an unmistakable but quiet torn-paper boundary."
|
||||
---
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue