lusipad

plcopen 对抗性探测轮——系统性猎杀未声明缺陷（KB-051 方法论流程化）。Use when a milestone closes, when the maintainer questions a capability claim, or when auditing semantic parity across domains.

plcopen 提交前门禁——全量测试、RT 扫描、回放校验的准确命令与已知误报坑。Use before any commit/push in this repo, when the user says 门禁/gates, or when another skill finishes an implementation step.

plcopen 提交信息格式——angular 类型 + 中文四段正文 + KB/声明变更标注 + AI 声明。Use when committing in this repository.

plcopen 出处纪律——禁止来源清单与外部代码/文档引入规则。Use before referencing external implementations, vendoring third-party content, or writing planner/OTG/blending algorithms.

plcopen 新特性/新语义的规格先行全流程（语义矩阵 → 人批 → 测试先行 → 门禁 → 文档同步）。Use when adding any new capability, FB semantics, planner behavior, or when the user proposes a feature touching core/.

Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes — Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/PRD asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".

plcopen harness 与文档体系的路由索引。

plcopen 已知边界（KB）登记流程——新核相对标准/旧线的行为边界如何编号与记录。Use when a change declares a behavior boundary, a deliberate deviation, or an explicit unsupported combination.

Use for Azure DevOps Server or on-prem TFS/Azure DevOps REST work when Codex needs to inspect projects, repositories, pull requests, work items, WIQL queries, wikis, search routes, test plans, test runs, test results, builds, or conditionally releases on a collection URL. Triggers include "Azure DevOps Server", "on-prem Azure DevOps", "TFS", collection-based REST calls, Windows-auth/PAT automation, and cases where the official Azure DevOps MCP for Azure DevOps Services is not the right fit.

当用户主要在查 Azure DevOps 的工单、PR、Pipeline 状态时使用。

当用户明确要求审查或总结某个 PR 时使用。

当用户要生成可复制、可转发、可提交的内容时使用。

当用户主要在查本地仓库代码、配置和实现逻辑时使用。

当请求来自定时任务时使用，输出稳定、可追踪的播报内容。

基于已有 PR 反馈继续在当前分支修复，而不是新开无关任务链。

分析 CI 失败并判断是否可安全自动修复，否则明确升级人工。

分析归一化后的 Azure DevOps 任务，并为 harness flow 生成结构化执行计划。

将批准后的任务计划转换为 ACP 编码请求，并收集执行结果。

Kickoff (开工) — run before starting any non-trivial task. Diagnoses the gap between what the user asked for (the map) and what the codebase or domain actually requires (the territory), then applies only the techniques needed — blindspot briefing for unfamiliar territory, premise challenges that falsify what's treated as fact, throwaway prototypes and feasibility spikes for "I'll know it when I see it", one-question-at-a-time interviews for ambiguities, reference extraction, and an unknowns-first plan. Use when the user says "kickoff", "开工", "帮我开始", "找找我的盲区", "help me start", "find my unknowns", "what am I missing", "I don't know where to start", or when a previous attempt — theirs or an AI's — came back wrong and they don't know why.

Quiz-me (考考我) — after a large change, or any change the user didn't author themselves (AI-written code), generates a report explaining the context, intuition, and mechanics of what changed — then quizzes the user on it and grades the answers. Recommends merging only when the user passes completely. Use when the user says "quiz me", "考考我", "测测我", "我不放心这次改动", "do I actually understand this change", "explain what happened then test me", or after a long working session before merge.

Wrapup (收工) — run after implementation work is done, before merging or sharing. Reads the diff, plan, and implementation notes, then produces what the moment needs — a buy-in document that leads with the demo and answers reviewers' objections up front, and a quiz that verifies the user actually understands what changed, recommending merge only on a full pass. Ends by promoting session learnings into permanent context. Use when the user says "wrapup", "收工", "干完了", "I'm done", "package this up", "ready to merge", "get buy-in", or finishes a long working session.

Kickoff (开工) — run before starting any non-trivial task. Diagnoses the gap between what the user asked for (the map) and what the codebase or domain actually requires (the territory), then applies only the techniques needed — blindspot briefing for unfamiliar territory, premise challenges that falsify what's treated as fact, throwaway prototypes and feasibility spikes for "I'll know it when I see it", one-question-at-a-time interviews for ambiguities, reference extraction, and an unknowns-first plan. Use when the user says "kickoff", "开工", "帮我开始", "找找我的盲区", "help me start", "find my unknowns", "what am I missing", "I don't know where to start", or when a previous AI attempt came back wrong and they don't know why.

Wrapup (收工) — run after implementation work is done, before merging or sharing. Reads the diff, plan, and implementation notes, then produces what the moment needs — a buy-in document that leads with the demo and answers reviewers' objections up front, and a quiz that verifies the user actually understands what changed, recommending merge only on a full pass. Ends by promoting session learnings into permanent context. Use when the user says "wrapup", "收工", "干完了", "I'm done", "package this up", "ready to merge", "get buy-in", or finishes a long working session.

Quiz-me (考考我) — after a large change, generates a report explaining the context, intuition, and mechanics of what changed — then quizzes the user on it and grades the answers. Recommends merging only when the user passes completely. Use when the user says "quiz me", "考考我", "测测我", "我不放心这次改动", "do I actually understand this change", "explain what happened then test me", or after a long working session before merge.

Use for Azure DevOps Server or on-prem TFS/Azure DevOps REST work when Codex needs to inspect projects, repositories, pull requests, work items, WIQL queries, wikis, search routes, test plans, test runs, test results, builds, or conditionally releases on a collection URL. Triggers include "Azure DevOps Server", "on-prem Azure DevOps", "TFS", collection-based REST calls, Windows-auth/PAT automation, and cases where the official Azure DevOps MCP for Azure DevOps Services is not the right fit.

一个会持续学习的 AI 视频导演技能。适用于 Codex 需要判断一条视频该怎么被做出来，并将其编排成剧本、分镜、素材清单、提示词和执行操作包的场景。