Research & Data

基于当天 Claude Code / Codex 活跃会话生成每日工作总结。扫描 ~/.claude/projects 下的 Claude Code .jsonl，以及 ~/.codex/state_5.sqlite 索引到的 Codex rollout-*.jsonl，从用户消息、助手回复、工具调用和命令结果中还原工作主题，以「全天时间线」为主轴（含甘特图把几点在做哪个项目画出来）串联整天工作、以「各项目推进」为副轴讲清每个项目，直接组织成 HTML 文档（不生成中间 md），并通过 lark-cli 发送到本人飞书邮箱。当用户说「总结今天的工作」「根据对话整理日报」「输出每日总结」或类似诉求时使用。

Deep-analyze a specific related / competing / concurrent paper — reading the ACTUAL paper, not its abstract — to decide the "isn't this just X?" question before a reviewer asks it: what it really does, whether it's prior or concurrent work, the strongest scoop a reviewer could claim, whether your delta is REAL or merely cosmetic, and the exact cite + positioning clause (or a scoop escalation) to apply. Saves the analysis to the paper's research folder and surfaces action points. Use whenever a paper close enough to yours surfaces — via verify-citations' nearest-neighbor, a reviewer's "this was already done", your own lit search, or an arXiv alert. Distinct from verify-citations (confirms a cite is real + one-line delta) and study-accepted-papers (mines a venue's accepted corpus for levers); this one does the deep per-competitor overlap/scoop analysis. Compose with verify-citations (feeds it neighbors), harden-paper (its nearest-neighbor-scoop axis calls this), and extend-paper (a scooped angle can become the

Comprehensive legal document review specialist for contracts, litigation documents, and real estate agreements — summarizing documents, flagging risk clauses, comparing contract versions, and checking compliance across any law firm size or practice area

Universal real-time CLI HUD & Statusline for Hermes, Claude Code, and Codex.

Use this skill for scene structure and game flow in melonJS — Stage subclasses, the state manager, transitions, pausing, the update loop, timers and tweens. Covers onResetEvent versus the constructor, state.set before state.change, the camera and world lifecycle per stage, and pause-aware timing. Triggers on: Stage, state, state.set, state.change, state.transition, state.pause, onResetEvent, onDestroyEvent, onActivateEvent, onDeactivateEvent, PLAY, MENU, LOADING, update loop, GAME_UPDATE, timer, setTimeout, setInterval, Tween, freeze.

Universal real-time CLI HUD & Statusline for Hermes, Claude Code, and Codex.

Reverse-engineer features from ELF binaries. Extracts CLI commands, state machines, protocols, Zod schemas, and data models. Automatically generates a /create-walkthrough prosecution brief with Mermaid diagrams. Uses /treesitter for AST analysis of bundled JS/TS source.

Search arXiv for papers and extract knowledge into memory. Use `search` to find papers, `learn` to extract knowledge.

Flexible data science analytics for any dataset. Auto-discovers schema, recommends charts, exports to create-figure. Works with JSONL, JSON, CSV from any source.

Multi-persona structured debate orchestrator. Personas research via /dogpile, consult colleagues via /ask, and argue toward nuanced synthesis on complex questions.

Anonymize supported CSV, JSON, UTF-8 text, and SQLite files using an explicit policy through the oai-trial project. Use for anonymize data, pseudonymize exports, redact policy literals, or discover and explicitly approve fuzzy name aliases. The skill is a thin CLI/Docker interface, not another engine.

Investigate whether a project (or every repo on the machine) is exposed to known and actively-exploited dependency vulnerabilities. Combines a deterministic per-ecosystem audit (npm audit, govulncheck, pip-audit, osv-scanner) with live web research into recent/active supply-chain incidents, then checks each project's lockfiles directly — the leading-indicator check that `npm audit` misses. Use when the user asks to "investigate security", "check for vulnerable dependencies", "are we affected by <a named CVE / package compromise>", "scan the projects for supply-chain issues", or wants a recurring security sweep.

Use when extracting specific data points from large agent output transcripts, kaizen analysis reports, or JSONL session files — tool timings, query counts, error summaries, or any structured facts — without loading raw data into orchestrator context. Activates when the orchestrator needs targeted facts from large files and context pollution must be avoided.

Transform transcript analysis findings into actionable improvements. Triggers on "generate hooks from findings", "improve agent", "fix anti-pattern", "kaizen improvement", "generate hook proposals", or "create improvement plan". Provides templates for hook generation, agent prompt refinement, skill patches, CLAUDE.md updates, and script automation from analysis data.

Diagnose and fix an unreliable, expensive, or unsafe LLM agent by auditing its harness (golden set, judge, cost caps, data layer, action approvals, tracing) before blaming or swapping the model. Use when someone says an agent is "burning tokens", "hallucinating", "brittle", gives inconsistent answers, asks whether to switch to a cheaper/better model, or wants to ship an agent or prompt change to customers.

Look things up before you code. Use for features, bug fixes, or anything that touches more than trivial code.

不良资产这行的钱，赚在**买入价**上——买贵了，后面怎么处置都白搭。

你给 AI 助手派了个重活，然后发现：发消息它不回、想插话只能排队、等十几分钟才回一句——等它抬头，需求早变了。根因不是模型慢，是主脑把重活放在前台跑，整个回合被占死。这个 Skill 把这套「重活外派 + 随时应答」机制做成可落地的东西：① 规则块（一页写清判级、派完立刻让出对话、工单四要素、只回收结论、自报≠事实、拍板留主脑、跨会话任务换机制）② 跨平台落地脚本 provision_agent_responsiveness.py：探测本机已有的 Agent 环境（Hermes 的 SOUL 置顶铁律块、Claude Code 的全局指令文件，这两条已实测），把规则块幂等写入（自动备份原文件、重复执行不重复写、绝不覆盖你已有内容）；其余平台（Codex / Cursor / WorkBuddy / 任意 Agent）则导出一份可直接粘贴的规则块 markdown，你粘到系统提示或自定义指令里就生效 ③ 体检脚本 audit_agent_responsiveness.py：逐项核对规则块是否存在、是否被后续改动覆盖、探测到哪些平台——没检测到就如实报「未检测到支持的平台」，不假报通过。纯标准库、零依赖、Windows/macOS/Linux 通用。触发词：AI不回话、等太久、插话卡住、助手响应慢、主脑被占住、重活外派、子代理、多智能体、响应速度、落地配置、系统提示。

存量财务报表在原表基础上改数字并保持三表勾稽关系。当用户要求「在原报表上改某些数字，其他数据跟着调整且格式完全不变」时使用：资产负债表/利润表/现金流量表的目标值调整、未分配利润与利润总额联动、勾稽校验。

想拿政府的钱，最容易白跑半年——层级选错、赛道不在目录里、返投算不清，材料递上去才发现方向不对。这套参谋按可执行顺序走：①先判层级（国家级/省级/市级/区级，门槛与目的完全不同）②赛道对目录：不在目录里直接如实说没戏，不让用户白跑③21 家基金卡片逐家对照（投资方式、返投比例、让利规则、申报条件、申报通道、真实案例），点名"你这个项目适合找谁"④返投博弈：算清落地要求，给"不想迁注册地"的三条合规路径⑤材料与时间表：申报材料清单、评审周期（6-12 个月是常态）与现金流提醒。与主技能 `funding-fit-diagnosis` 配套（做整体融资研判时用主技能，VC 与政府双轨分开算）。输入：项目基本情况与注册地/落地意愿；输出：层级判定 + 可申报基金名单 + 返投方案 + 材料清单 + 时间预期。不承诺拿到资金，申报须如实提交材料；政策口径标注截止时间，细则请向当地主管部门核实。触发词：政府基金、政府引导基金、产业基金、政府补贴、专项资金、项目申报、补贴申报、申报条件、返投、招商落地、母基金、投资方式、让利、国家级基金、省级引导基金、市级基金、科技成果转化、专精特新、科技型企业、中小企业发展基金。

公司出事，十有八九出在**治理和控制权**上——股权结构一开始搭歪，后面融资、上市、分家全是坑。

拿到 TS 或投资协议，看半天不知道哪条能签哪条得改——风险全藏在具体条目里，一句"整体没大问题"最要命。这套审查按 15 条核心条款逐条过：估值 Pre/Post、清算优先权、反稀释（完全棘轮是红线）、优先购买权、领售/拖售、对赌与业绩承诺、回购条款、个人连带担保、董事会与一票否决、保护性条款、竞业与知识产权、信息权、交割条件、股东会表决权、违约责任。每条给三级判定（🔴 红线不改就别签 / 🟡 可谈争取改 / 🟢 行业惯例）+ 谈判话术 + 替代方案，并区分"对公司"和"对创始人个人"——个人连带回购是绝大多数血案的源头，重点标红。另附谈判筹码排序、常见坑清单、BP 与数据室准备。与主技能 `funding-fit-diagnosis` 配套（做整体融资研判时用主技能）。输入：TS/投资协议条款（粘贴或描述）；输出：逐条判定表 + 必须改的清单 + 谈判话术 + 底线建议。**不构成法律意见**，正式签署与司法效力请找执业律师确认。触发词：TS、Term Sheet、投资条款清单、投资协议、增资协议、股权转让协议、对赌、业绩承诺、回购、清算优先权、反稀释、完全棘轮、个人连带、一票否决、领售权、拖售权、估值、Pre/Post、股权稀释、条款谈判、FA。

写完的稿子发布前想知道会不会被判 AI 生成、平台审核会不会卡、读者会不会一眼看出机器味——这套是检测到清洗的完整闭环。

多个 AI 助手同时干活时，浏览器会互相抢占：弹窗抢焦点、任务互相打断、你正在用的 Chrome/Edge 被顶掉——**你连自己的网页都点不动了**。