web-access

Plugin web-access from marketplace web-access in mykcs/myk-skills. Commands read from the repo's manifest — 1 month ago.

Skills (52)

• harness-upgrade

  Research-grounded workflow for upgrading Claude/Codex/agent harnesses. Refreshes current runtime state, searches recent primary research and official vendor guidance before design changes, minimizes hot context and tool surface, and verifies upgrades with independent evidence.

• host-self-evolve

  本地主机 Claude Code 协调 + 自我进化 (v3.3.1 cross-reference + v3.3.0 PER Workflow): 提升 ~/.claude/ 跨层一致性 + §I.4 8 步循环 + N-tool fan-out internalize. 5 Layer: Layer 0 5 commands gate / Layer 1 6 sub-task audit / Layer 2 cleanup orphan / Layer 3 N-tool fan-out / Layer A.2-A.4 5 字段自检 + 4 站 CI gate. 触发词: 主机自升级, /h

• phd-scout

  PhD Advisor Intelligence Gatherer — 申博情报搜集系统。自动抓取清北复交浙 AI/LLM/Agent 方向导师信息，判定危险信号，写入飞书。 Use this skill whenever the user asks to find, scout, research, or gather information about PhD advisors or mentors in Chinese universities (清华/北大/复旦/上交/浙大), especially for AI/ML/LLM/Agent research directions. Th

• web-design-engineer

  Build high-quality visual Web artifacts using HTML/CSS/JavaScript/React — web pages, landing pages, dashboards, interactive prototypes, HTML slide decks, animated demos, UI mockups, data visualizations, and more. Use this skill whenever the user's request involves a visual, interactive, or front-end

• webapp-testing

  Use when testing local web apps with Playwright, including screenshots and UI debugging.

• weekly-report-phd-ARCHIVED

  ⚠️ ARCHIVED 2026-07-16 — Notion block 布局 + verify 协议位已迁 paper-into-notion/references/notion-block-layout.md + notion-content-verify.md. 本文件保留作为历史档案 (周报项目实证), 不再触发. 项目级周报流程仍按 §A 7 件 checklist, 但走 paper-into-notion 主 SKILL.

• agent-knowledge-garden

  Audit and maintain Agent-facing repository knowledge across long-lived projects: keep root instructions small, preserve one writable owner per fact, detect stale/current conflicts, repair docs/agents navigation, and update the account registry without forcing one technical architecture onto every re

• anysearch

  Real-time search engine supporting web search, vertical domain search (23 domains), parallel batch search, and URL content extraction.

• auto-feishu-digest

  自动追踪领域重要 articles + tech blogs → 按 7 维评分规整 → 自动推送飞书表格。MVP 跑 **N-tool fan-out (N 当前 = 6, per [~/.claude/rules/protocols/N-tool-search.md](https://example.invalid/~/.claude/rules/protocols/N-tool-search.md) v1.1.2)** × 3 大类信息源 (论文会议 / 国内博客 / 国外博主) × daily 浅 + weekly 深双轨。立 loop-engineering skill 子方向字段

• backup-claude-settings

  Backup Claude Code settings.json and .claude.json to a GitHub repository with timestamped filenames. Automatically detects the active API provider (Kimi, MiniMax, Anthropic, etc.) from ANTHROPIC_BASE_URL and uses it as the filename prefix. Commits and pushes immediately. Includes mandatory diff audi

• code-review

  Reviews uncommitted changes for security vulnerabilities, code-quality issues, and best-practice violations, then blocks the commit if CRITICAL or HIGH issues are present.

• content2html

  内容 (arxiv paper / DOI / PDF / 本地 progress) → HTML (slide + long-form) 4 产物自动生成。 Fork guizang-ppt-skill Style B 视觉 (IKB 蓝 / Swiss editorial)。 独立 Astro project，部署到 mykcs.github.io/content2html/。 双语支持，默认中文 (zh)，可切英文 (en)。 触发场景：用户问 "制作组会 slide" / "arxiv paperlink 转 html" / "周报 slide" / "论文 long-form 摘要"

• context-budget

  Cross-harness context-engineering workflow for reducing hot-context waste, improving just-in-time retrieval, preserving structured long-horizon state, and detecting context/tool overload without assuming a fixed context-window size.

• conversation-closeout

  End-of-conversation knowledge closeout. Use when the user says the conversation is ending, wants lessons/preferences/memories/workflows preserved, asks to summarize and sediment what was learned, or wants reusable engineering experience written back so future work is faster. Sweep the accessible con

• docs

  Looks up current documentation for a library or topic via Context7 MCP (resolve-library-id + query-docs) and returns a summarized answer with code snippets.

• domain-modeling

  Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.

• fresh-agent-acceptance

  Black-box acceptance for completed work when the originating conversation may have a context advantage. Use whenever the user says they fixed/built/changed something but is not sure it is really solved for a fresh Agent, wants a short prompt to copy into brand-new windows, wants a no-context/hidden

• friction-mining

  摩擦挖掘闭环：扫最近 session 转录里的摩擦事件（误解请求 / 错误方案 / 改动过多 / 啰嗦格式）， 对复发 ≥3 次的模式自动起草硬规则 + 验证探针，开原子 PR。把一次性手动挖掘固化成可重复的一键动作。 触发场景: 用户说 "摩擦挖掘 / friction-mining / 挖摩擦 / 扫 session 找复发问题 / 自愈闭环 / friction"， 或想把近期反复踩的坑固化成规则 + 探针。 Trigger when user says: "摩擦挖掘" / "friction-mining" / "挖一下最近的摩擦" / "扫 session 找复发问题" / "自愈闭

• gdkvm-experiment-workflow

  Orchestrate reproducible GDKVM and related GPU experiments from protocol freeze through launch, monitoring, recovery, verification, and evidence closeout. Use for whs512/rtx6 (4×RTX 3090), or for lyg2171 (daemon label) via wangrui_root → root@wangr-dev (8×RTX 5090); status-only GPU questions remain

• gdkvm-train-launcher

  GDKVM 训练启动 skill（2026-06-30 立, v1.0）— 在六号服务器（202/rtx6, 172.31.71.202）跑 hydra + torchrun 双卡训练。一行连接 + systemd-run 启动 + journalctl 监控 + GPU/CUDA 0,1 隔离 + loginctl Linger 必备 + hydra config 必须在 config/ 目录。触发场景：用户说「六号服务器」「202 (rtx6)」「两张卡」「GDKVM 训练」「code_modern_v3」「systemd-run 启动」时必读。

• grill-with-docs

  A relentless interview to sharpen a plan or design, which also creates docs (ADR's and glossary) as we go.

• grilling

  Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.

• harness-audit

  Runs a deterministic repository harness audit and returns a prioritized scorecard across 7 fixed categories (tool coverage, context efficiency, etc.).

• healer-cannot-self-heal

  医者不可自医 — Session 急诊室。claudecode 在判断当前对话已失控（上下文耗尽、反复卡同一处、漂移）时， 不修复自己，只输出一份"密集证据 + 最小化怀疑标注"的症状报告，让用户（医者）做判断。 核心原则：医者不可信 → 输出保持怀疑语气，处方显式标 LOW-CONFIDENCE。 触发：用户判断 session 不行时主动召唤（不是自动 hook）。 与 session-chapter 互斥：session-chapter 是"搬家"，本 skill 是"急诊"。 反 trigger（绝不接管）：Claude Code 内置 `/doctor`（plugin/MCP 诊断器

• memory-bench

  Benchmark claudecode's memory system against official SOTA (LongMemEval / LOCOMO / MSC) + Oracle + ablation 5-deletion. Tests 4 metrics: recall (35%) / cross-source consistency (25%) / compliance (30%) / token economy (10%). Runs opus-as-judge over 50 recall questions + 15 consistency checks + 12 co

• kimi-webbridge

  Kimi WebBridge lets AI control the user's real browser — navigate, click, type, read, screenshot, and interact with any website using the user's actual login sessions. Use this skill whenever the user wants to interact with websites, automate browser tasks, scrape web content, or perform any action

• lark-base

  飞书多维表格（Base）操作：建表、字段、记录、视图、统计、公式/lookup、表单、仪表盘、workflow、角色权限；遇到 Base/多维表格/bitable 或 /base/ 链接时使用。文件导入转 lark-drive，认证/授权转 lark-shared。

• lark-doc

  飞书云文档（Docx / Wiki 文档）：读取和编辑飞书文档内容。当用户给出文档 URL 或 token，或需要查看、创建、编辑文档、插入或下载文档图片附件时使用。文档中嵌入的电子表格、多维表格、画板，先用本 skill 提取 token 再切到对应 skill。当用户给出 doubao.com 的 /docx/ 或 /wiki/ URL/token 时，也应直接使用本 skill；路由依据是 URL 路径模式和 token，而不是域名。不负责文档评论管理，也不负责表格或 Base 的数据操作。当用户明确要操作飞书思维笔记时，也使用本 skill。

• lark-drive

  飞书云空间（云盘/云存储）：管理 Drive 文件和文件夹，包含上传/下载、创建文件夹、复制/移动/删除、查看元数据、评论/权限/订阅、标题、版本和本地文件导入。用户需要整理云盘目录、处理云空间资源 URL/token、判断链接类型/真实 token/标题，或导入 Word/Markdown/Excel/CSV/PPTX/.base 为 docx/sheet/bitable/slides 时使用；doubao.com 云空间 URL/token 也按资源路径和 token 路由，不回退 WebFetch。不负责：文档内容编辑（走 lark-doc）、表格/Base 表内数据操作（走 lark-

• lark-wiki

  飞书知识库：管理知识空间、空间成员和文档节点。创建和查询知识空间、查看和管理空间成员、管理节点层级结构、在知识库中组织文档和快捷方式。当用户需要在知识库中查找或创建文档、浏览知识空间结构、查看或管理空间成员、移动或复制节点时使用。当用户给出 doubao.com 的 /wiki/ URL/token 时，也应直接使用本 skill，不要因为域名不是飞书而回退到 WebFetch；路由依据是 URL 路径模式和 token，而不是域名。不负责：上传文件到知识库节点下（走 lark-drive）、编辑文档/表格/Base 内容（走 lark-doc / lark-sheets / lark-bas

• loop-engineering

  pipeline = 决策树 + 4-stage 串行 + 自决矩阵 + 5 字段验收 把"调研一个领域 / 沉淀一份 paper / 跑完一组 8 决策"这种有终点、有阶段、有可验证交付物的任务, 变成可复用 SOP。 version: "1.0" author: mykcs category: research - loop-engineering - /loop-engineering - pipeline - research pipeline - decision pipeline - 调研流水线 - 串行 sub-agent

• mcp-builder

  Use when building MCP servers to integrate external APIs or services.

• mmx-cli

  Use mmx to generate text, images, video, speech, and music via the MiniMax AI platform. Use when the user wants to create media content, chat with MiniMax models, perform web search, or manage MiniMax API resources from the terminal.

• omc-reference

  OMC agent catalog, available tools, team pipeline routing, commit protocol, and skills registry. Auto-loads when delegating to agents, using OMC tools, orchestrating teams, making commits, or invoking skills.

• paper-into-notion

  URL → 自动填 3 字段 (页面 / 状态 / 模态类型) 到 Notion `论文` database. multi_select (教育类型/关键词/标签) + rich_text 亮点 **永不覆盖**已有值 (PATCH body 只含 select/title). 适用 arXiv / 公众号 / 博客 / Twitter / GitHub / bilibili / youtube / 小红书 / 知乎. 11 scripts + 6 templates + 6 references. weiying20260624 PhD 申请场景. v4.7 (2026-07-25): bo

• post-pr-merge-ff-verify

  PR 合并 (gh pr merge) 之后必跑的 fast-forward verify SOP: ahead/behind 检查、 local-only commits 兜底、reset --mixed 撤回路径。当用户合并 PR、提到 ff verify、 ahead/behind 不为 0 0、或需要撤回本地 merge commit 时触发。 version: 1.0.0 category: workflow author: mykcs migrated-from: ~/.claude/rules/post-pr-merge-ff-verify-rule.md (2026-07-25

• project-report

  制作或审计面向老师、领导、学长、同实验室/同组织同事的项目与科研汇报。只要用户说“制作汇报”“写汇报”“改汇报”“Notion 汇报”“网页汇报”“给老师看进展”“项目进展总结”“研究进展汇报”或要求把已有实验/项目结果整理成可读汇报，就优先使用这个 skill。目标不是写流水账，而是让“知道这个项目、但不了解全部细节”的读者在最短时间看懂：已经做了什么、实际得到什么、哪些还没做、这些结果把下一步实验方向推向哪里。支持 Notion、网页、Markdown、飞书等内容载体，并包含生成与整页审计/修复两种模式。 不用于 PhD 导师画像/申博调研（用 teacher-report），也不用于单篇

• record-case

  统一知识归档 skill：先输出位置，再决策类型，最后按模板深度归档。 支持 Case(深度)/Pattern(轻量)/Decision(决策) 三种类型，覆盖所有知识沉淀场景。 触发场景: 用户说"记录这个/归档/留档/case 一下/学到了/总结经验"、修复完成后的知识沉淀、调试 ≥2 轮后的根因落地。 version: "3.0.0" author: mykcs category: knowledge triggers: - 记录这个 - 归档 - 留档 - case 一下 - 存为案例 - record case - archive this - 写成 case - 学到了 - 记录经

• report-making

  “制作汇报”——创建、重写或审计面向老师、领导、同实验室/同组织成员的 Notion 汇报、网页汇报、周报、研究进展页或项目总结。 这些读者通常知道项目大概在做什么，但不了解最新细节。每当用户说“制作汇报”“做/改 Notion 汇报”“网页汇报”“给老师/领导看” “介绍最近做了什么”“检查这篇汇报”，都应优先使用本 Skill。按重要度从高到低逐项检查：先保证事实与科研结论，再保证首屏能看懂 “做了什么 / 发现了什么 / 没做什么 / 下一步做什么”，最后再处理术语、版式和润色。无论创建还是修改，最终聊天回复必须给出 ✅ 已完成 / ❌ 未完成的验收清单。 version: "1.0.0

• rules-distill

  Distills rules by reading project rules, cross-referencing, and emitting a verdict (keep / consolidate / deprecate / move). Operates per the rules-distill inventory + cross-read + verdict workflow.

• server-artifact-governance

  Organize and reclaim space on authorized research/ML servers without losing scientific value or touching other users' assets. Use when the owner asks to 整理服务器, 清理 checkpoint/adapter/run, 腾空间, 迁移/归档科研资产, 清理大型文件, audit storage, or decide what can be deleted. Resolve the target server's current authori

• skill-creator

  Use when creating, editing, or optimizing skills, or running skill evals and benchmarks.

• skill-evolution

  周自演化 skill — 读 friction 数据起草 SKILL.md v-bump + 沙箱 A/B test + 仅胜出 ship

• behavioral-skills-management

  Skills 目录 symlink 矩阵 + git 同步规则（2026-06-01 重构）

• teacher-report

  Generate OR audit a PhD advisor / professor intelligence dossier as a Feishu wiki doc (Docx). **Generate**: user mentions researcher/advisor/老师/导师 + asks 调研/写一份报告/整理材料 → 4-section report (1.1 自评 / 1.2 导师画像 (1.2.1-1.2.5) / 1.3 论文全景 (1.3.A 顶会 10 v0.3.0 增强 h3 + 1.3.B 主题表 5-7) / 1.4 数据来源) + 1.5 套磁准备清单.

• test-gated-debug

  测试门控迭代调试：先写一个会失败的最小复现测试，再对着它迭代到转绿，自审后才提交。 治「修了一个又弄坏另一个」的打地鼠循环。适用于代码 bug 修复，不适用 settings/config 修复。 触发场景: 用户说 "测试门控 / test-gated / 先写失败测试 / 严格调试 / 收敛调试 / test-gated-debug"， 或代码 bug 反复修不好、改 A 坏 B、需要可证伪的根因验证。 Trigger when user says: "测试门控" / "test-gated" / "先写失败测试再修" / "严格调试这个问题" / "收敛调试" / "test-gated

• verifier-pass2

  Two-pass subagent validation. Pass-1 (Code Review agent) flags findings; Pass-2 (this skill) validates each finding adversarially with high confidence before action. Two-pass pattern modeled after Anthropic's official code-review plugin. Triggered by: pass-2 verify, validate review, second pass, adv

• verify

  Canonical cross-harness verification workflow for build/type/lint/test/security checks, targeted quality gates, adversarial review, pass-2 finding validation, and high-impact verification portfolios that separate implementation failures from infrastructure noise.

• website-improve

  Evidence-first website improvement workflow (v4.3.0). Uses three independent Planner → Executor → Verifier roles for full runs, plus a proportional situational lens that makes every human-facing web change consider semantic HTML, mental models, information density, and reading flow without forcing h

• site-modernizer-workspace

  Working directory for site modernization iterations

• workflow-evolution

  Evolve engineering workflows, CI/resource usage, Agent harnesses, host control planes and knowledge hierarchy. Use for workflow evolution, broad modernization, CI quota/cost reassessment, cross-project adoption, harness upgrades, knowledge buoyancy / case distillation or host self-evolution. Refresh

• xiao-de

  小德的经济学课堂 — 用故事和经济学原理解释「为什么明知该改变却做不到」。 触发场景：小德、改变自己、鸡汤、贫穷循环、努力、延迟满足、跳出循环。 version: "1.0.0" author: mykcs category: persona triggers: - 小德 - xiao-de - 改变自己 - 鸡汤 - 贫穷循环 - 延迟满足 - 跳出循环 tags: - persona - economics - psychology - motivation

Similar plugins