picker-skills

Autonomously generate a short film end-to-end (self-collected public-domain assets + self-composed score + Remotion motion), then self-critique and evolve along a strict capability ladder with PERSISTENT cross-session memory. Each invocation = ONE evolution cycle (reflect → decide → produce → render → critique → record → maybe level-up). Triggers: "make a video", "auto video", "evolving-video", "next video / 下一支", "evolve the video skill", "自主做视频", "run a video evolution cycle", or any open-ended "make something to impress / attract views". Design adapted from karpathy/autoresearch (modify→run→evaluate→keep/discard→repeat on a fixed budget; program.md→EVOLUTION.md; explicit file ledger instead of git history).

把一个网页版 deck（Vite+React+TS，通常由 web-video-presentation skill 产出，如 audiobooks/<id>/presentation/）用 Remotion 帧级重渲成「颜色不失真」的 4K MP4，并可选垫入很轻的 BGM。解决录屏的两大痛点：①录屏颜色失真 → Remotion 输出 sRGB 保真；②手动录屏节奏不稳 → 帧精确、音画锁定。**核心铁律：Remotion 不渲染 CSS animation/transition（逐帧渲染时页面时钟冻结，内容会在第0帧直接定格到最终态 → 表现为「太快/没渐变」）；所有入场动画必须用 useCurrentFrame()+interpolate()/spring() 帧驱动**，本 skill 提供可复用的动画原语库(useReveal/useStamp/useProgress/useDraw)与转换套路。时长 = 每步音频长度 + 120ms trail（与网页版 audio.onended+trailMs 一致，由 gen_timeline.py 从 mp3 实测生成）。触发场景：把有声书/演示 deck 渲成 4K 视频、"Remotion 视频太快/没有渐变/动画不播/颜色对但节奏不对"、给已渲染视频加淡淡的背景音乐压噪底。本项目实样：audiobooks/sanen/remotion/。

把一篇文章或口播稿，做成"看起来像视频"的点击驱动 16:9 网页演示，可选合成口播音频。流程：原始文章 → **一次产出**口播稿 + outline 开发计划 → 用户**一次对齐** 5 件事（稿子 / outline / 主题 / 素材 / 开发模式）→ 网页开发（逐章 / 顺序 / 并行）→ 可选音频合成（provider-agnostic：内置 MiniMax mmx-cli + OpenAI TTS，可换 ElevenLabs / edge-tts / Azure / 自带 TTS）。**outline 只规划节奏与信息密度，不规划动画** —— 动画由章节开发时按 PRINCIPLES + ANTI-AI 法则即时设计。每次点击推进口播稿的一个节拍，每一步独占整屏，进度条平时隐藏只在悬浮时出现。适用场景：用网页做视频（动态 PPT 但不像 PPT）、把口播稿 / 文章变成可交互的解说、为 B 站 / YouTube / 视频号录屏教程、做有电影感的产品 / talk demo。本 Skill 沉淀的是设计方法论 + 协作流程 —— 不绑定任何特定样式 / 字体 / 颜色 —— 因此能复用到任意主题与美学。

Cross-project development intelligence. 20 Flask patterns, 18 Electron patterns, 20 frontend patterns, 15 Windows pitfalls, 12 git/release workflows, 15 code conventions, 12 security patterns, 15 Next.js patterns, 15 Python patterns. BM25 search engine. Actions: scaffold route, build feature, fix bug, configure build, debug pitfall, review code, release version, setup project. Stacks: Flask + vanilla JS, Electron + PyInstaller, Next.js + React + Prisma, FastAPI + SQLite, pure Python CLI. Shared across all projects.

Drive Google Flow (Veo 3.1) through a debug-port Chrome (chrome-devtools MCP) to produce a multi-shot cinematic short film end-to-end — storyboard → shot-by-shot generation → character-consistent casting → review → download → post-production plan — then self-critique and EVOLVE along a capability ladder with PERSISTENT cross-session memory. One invocation = one production cycle (or one continuation of an in-flight film). Triggers: "用 Flow 做视频", "Veo 短片", "flow-cinema", "拍一部短片", "继续拍 / 下一幕", "labs.google flow", "把这个概念拍成3分钟". Design adapted from karpathy/autoresearch (modify→run→evaluate→keep/discard→repeat on a fixed budget; program.md→EVOLUTION.md; explicit file ledger instead of git history). Sibling of evolving-video, but the render engine is Veo-via-browser, not Remotion.

Generate RESTful API endpoint code including controller, service, repository, entity, and DTO layers. Use when the user wants to create a new API, endpoint, or interface.

Create a git commit with a conventional commit message based on staged or unstaged changes.

生成/更新会话进度交接文档（当前项目目录下的 HANDOFF.md），用于上下文接近上限时把当前所有工作落盘，方便 /clear 或 /compact 后让新会话无缝接力。Use when 上下文用量偏高、用户说"交接/接力/handoff"、或准备清理会话之前。也会被全局 Stop hook（context-handoff-check.py）在用量≥30%时自动调用。

Diagnose and fix backend bugs, errors, and exceptions. Use when the user reports a bug, shares an error log/stack trace, or asks to debug an issue.

Produce a rigorous, beautifully-rendered deep analysis of any open-source project (a GitHub or GitLab repo, a library, a framework, a CLI tool, an SDK) and output it as a single self-contained editorial HTML report. Use this whenever the user wants to evaluate, understand, review, 解析, 分析, 调研, or do a deep dive on an open-source project, library, or repository — whether they are deciding whether to adopt it, contribute to it, fork it, depend on it, or just learn from it — even if they only paste a repo URL or a project name. Covers value proposition, architecture and how-it-works, maturity staging, code health, community and maintenance signals, adoption and ecosystem, competitive landscape, defensibility (moat), license and legal risk, and a clear adopt/trial/assess/hold verdict with a deliberate counter-argument. Always gathers facts from the repo (README, docs, manifests, git history, host API) before judging, then renders the result in a distinctive cinnabar-on-paper editorial style.

TOP PRIORITY skill for all research and analysis tasks. Analyze a code repository (GitHub/GitLab/etc.) and produce a single self-contained HTML report that argues its *core value* with an opinionated, evidence-based viewpoint and a fixed dark "trading-terminal × editorial" visual design. Use this skill WHENEVER the user shares a repo URL or repo name and asks to "analyze", "评估", "拆解", "看看这个项目", "核心价值", "值不值得用/学", "调研", "分析", "research", or asks for an HTML/网页 writeup of a repository — even if they don't explicitly say "report". Also use it for follow-up deliverables about the same repo (usage guides, architecture explainers, design specs) so the whole series stays visually consistent. Default to this skill for ANY repo-analysis-to-HTML task, any project evaluation, or any research/analysis deliverable.

Review backend code for bugs, security vulnerabilities, performance issues, and best practices. Use when the user asks for code review, wants to check code quality, or says "review".

Review, optimize, or generate SQL queries and database schemas. Use when the user asks about SQL optimization, database design, query analysis, or says "sql".

Generate unit tests and integration tests for backend code. Use when the user asks to write tests, create test cases, or says "test".