skills

Search doctoral dissertations for a given academic field, compile professional sites and databases with description and update frequency, preserve complete search records, and independently verify every citation link before including in the final report. Use when user wants to explore a new research field, search for recent doctoral dissertations, find authoritative professional websites and databases for a discipline, or needs to verify and format academic citations with clickable links.

Academic paper writing assistant for top-tier conferences and journals with AIGC-aware natural voice and figure quality enforcement. Triggers when: Writing academic papers or articles, needing literature search and citation formatting, preparing manuscripts for conference submission, creating paper structure, generating SVG/PDF figures, drafting reviewer response letters, or reviewing/polishing to reduce AIGC detection rates. Commands: - /paper new <topic> - Start new paper with topic analysis - /paper search <keywords> - Search and collect literature - /paper cite - Format citations properly - /paper structure - Generate paper outline - /paper review - Review, polish, and reduce AIGC markers - /paper detect <text> - Analyze AIGC features in text - /paper zh <topic> - Write Chinese journal article (CCF/知网 format) - /paper figures - Generate publication-quality SVG/PDF figures - /paper rebuttal - Draft reviewer response letter Capabilities: AIGC-aware academic writing, natural academic voice, literature search from Google Scholar/arXiv/DBLP/Semantic Scholar, citation formatting (IEEE/ACM/APA/GB-T7714), paper structure generation, mathematical notation, Chinese academic publishing, SVG/PDF figure generation, reviewer response drafting, iterative review with AIGC rate reduction, opencode-specific collaborative writing patterns

Security review for skills before installation, detecting dangerous patterns and assessing risk levels. Triggers when: Installing a new skill, periodically reviewing installed skills, or suspecting security issues in a skill. Commands: - /安检 <技能路径> - Audit skill security - /安检 scan <路径> - Deep scan - /安检 list - List installed skill risks - /安检 fix <技能> - Fix security issues - /security <skill-path> - English command for skill security audit Capabilities: Static code analysis, dangerous pattern recognition, risk assessment, auto-fix, user interaction decisions

申报材料包（Application Bundle）：把一份"申报/上报需求"组装成套交付物——填写表格模板（合并单元格保格式）、生成正文材料（事迹/推荐书/申报书/报告正文）、匹配并嵌入证据附件图、统一命名、打包目录。泛化于评优推荐、职称申报、课题申报、验收材料、答辩材料、毕业设计材料等一切"表格+正文+证据"型材料组装。 Triggers when: 需要填申报/评优/验收表格并按模板出正文与证据材料（评优推荐表、优秀个人/先进集体、职称申报表、课题申报书、验收报告、答辩材料）、把"填好的表+事迹/申报正文+证书/证明图片"组装成一套可上交的材料包。 Commands: - /材料包 <需求> - 组装整套申报材料（表格+正文+证据） - /材料包 填表 <模板.docx> <数据> - 填写表格模板（合并单元格保格式） - /材料包 正文 <内容> - 生成事迹/申报正文（docx，宋体小四/标题规范） - /材料包 证据 <清单> - 匹配并嵌入证据附件图（PDF→PNG转换） - /材料包 打包 <目录> - 按命名规范整理成材料包目录 - /bundle <req> - English alias Capabilities: docx合并单元格表格定位与填写（保格式、保留签字行）、正文 docx 生成（标题黑体小三/正文宋体小四）、证据图收集与嵌入（证书PDF转PNG、每图加注说明）、材料包目录组织与命名、格式保真校验

Pre-publish review with multi-layer deep analysis for code quality assurance before release. Now with AI attribution detection (ODS-inspired), risk-weighted scoring, and review-tier routing. Triggers when: Preparing to publish an npm package, needing pre-release review, or checking code change quality. Commands: - /把关 - Start full pre-publish review - /把关 check - Check unpublished changes - /把关 version - Suggest version bump - /把关 report - Generate review report - /把关 attribution - Show AI attribution summary - /review <task> - English command for pre-publish review Capabilities: Detect unpublished changes, AI attribution detection (Co-Authored-By/Assisted-by trailers), per-change deep analysis, multi-role review (architect/developer/tester/security/docs), risk-weighted scoring, review-tier routing (auto/standard/elevated), version suggestion, release risk assessment

Enrollment guidance assistant that searches official data for cross-referenced analysis and practical advice on college entrance and graduate school applications. Triggers when: Needing enrollment data analysis, checking admission scores, analyzing major employment prospects, or researching school admission statistics. Commands: - /报考 <查询内容> - Search and analyze enrollment data - /专业 <专业名> - Analyze major employment prospects - /学校 <学校名> - Analyze school admission data - /一分一段 <省份年份> - Query score ranking tables - /分数线 <学校专业> - Query admission score cutoffs - /enroll <query> - Search and analyze enrollment data (English) - /major <name> - Analyze major employment prospects (English) - /school <name> - Analyze school admission data (English) Capabilities: Official data search from education authorities, cross-referenced data verification, historical trend analysis, score-ranking conversion, admission probability estimation, employment prospect analysis

批量文件管道（Batch File Pipeline）：对一批文件执行可复用的处理管道——方向检查与校正（OCR检测旋转）、按命名规范统一重命名、格式转换、去重、校验。泛化于一切"一批文件要统一处理整理"的任务：扫描证书、照片、下载的媒体（音频/视频）、PDF文档、字幕、图片。防丢失是硬约束：先验证新文件再动旧文件。 Triggers when: 需要批量旋转/校正一批图片或扫描件、按统一命名规范批量重命名文件、批量转换格式（PDF↔图片、docx↔md、媒体压缩）、去重、检查一批文件是否规范、为一批文件生成清单。 Commands: - /批量文件 <目录> rotate - 批量检测并校正图片旋转方向（OCR盒坐标判断90°/180°） - /批量文件 <目录> rename <规范> - 按命名规范批量重命名（支持模板/正则/顺序号） - /批量文件 <目录> convert <目标格式> - 批量转换格式 - /批量文件 <目录> dedupe - 按内容哈希去重 - /批量文件 <目录> verify <规范> - 校验文件完整性与命名 - /batchfiles <dir> <action> - English alias Capabilities: OCR盒坐标旋转检测（竖排文字比例判断90°旋转+CW/CCW判定）、PIL旋转校正（quality=95）、按模板批量重命名、格式转换管道、SHA-256内容去重、先验证后删旧的防丢失协议、批量清单生成、命名规范正则校验

Writing skill that mimics user blog style with concise, colloquial, and sincere language while avoiding AI-detectable patterns. Triggers when: Needing to mimic user writing style, write concisely, compose blog articles, or create technical documentation. Commands: - /简写 <主题或文本> - Write or rewrite in user's style - /write <topic/text> - Write in user's style (English) - /风格检查 <文本> - Check writing style compliance - /style-check <text> - Check writing style compliance (English) Capabilities: User blog style imitation, concise colloquial expression, AI-pattern avoidance, style compliance checking, bilingual Chinese-English support

布阵 - Multi-agent orchestration for opencode. Ships 8 REAL opencode agents (bz-orchestrator primary + bz-explorer/bz-oracle/bz-council/bz-librarian/bz-designer/bz-fixer/bz-observer subagents) with per-agent models, temperatures, and config-level permission enforcement. 3 presets (opencode-free/deepseek/mixed) switchable via scripts/apply_preset.py. Safety enforced at opencode permission layer plus prompt-layer rules (budget caps, doom-loop, stale-work, proto-pollution, model fallback, 9-type delegation errors). Includes anti-duplication, 6-section delegation prompt, auto-continue, notepad, boulder completion report. Triggers on non-trivial multi-step coding tasks needing delegation, parallel exploration and review, or multi-phase refactors with gates. Inspired by code-yeongyu/oh-my-openagent (real original, SUL) and alvinunreal/oh-my-opencode-slim (MIT); all prompts and rules are original Chinese rewrites.

Core patterns for AI coding agents based on analysis of Claude Code, Codex, Cline, Aider, OpenCode. Triggers when: Building an AI coding agent or assistant, implementing tool-calling loops, managing context windows for LLMs, setting up agent memory or skill systems, or designing multi-provider LLM abstraction. Commands: - /agent-patterns - View coding agent patterns and architecture (informational skill) Capabilities: Core agent loop with while(true) and tool execution, context management with pruning and compression and repo maps, tool safety with sandboxing and approval flows and doom loop detection, multi-provider abstraction with unified API for different LLMs, memory systems with project rules and auto-memory and skill loading, session persistence with SQLite vs JSONL patterns

Converts .docx documents (manuals, textbooks, reference books) into self-contained, searchable, mobile-friendly HTML knowledge bases with embedded images, chapter navigation, and full-text search. Triggers when: User provides a .docx file and requests a knowledge base, searchable HTML, mobile-friendly reader, or a generic knowledge-base JSON. Commands: - /知识库 <docx路径> [输出目录] - Full pipeline: extract → clean → generate HTML + JSON + images - /知识库 extract <docx路径> - Extract text + images from docx, save to structured JSON - /知识库 clean <json路径> - Clean and polish the extracted JSON (spacing, headers, captions) - /知识库 generate <json路径> - Generate standalone HTML with search + nav + embedded images - /知识库 verify <html路径> - Run automated quality checks (images, spacing, search function) Capabilities: docx text+image+table extraction, CJK spacing normalization, stray header removal, image caption grouping, English diagram label separation, data-URI embedding for single-file output, full-text search with result excerpts, mobile-responsive layout, dark mode support

Pure-Python legacy Word (.doc) → .docx converter. Parses the Word 97/2000/2002/2003 binary format (and the Word-97-compatible output written by WPS Office) directly — no LibreOffice, no Microsoft Word, no textutil, no antiword/catdoc, no external binaries of any kind. Only olefile + python-docx + lxml. Triggers when: Converting a .doc file to .docx while preserving text, fonts (incl. 宋体/黑体/仿宋 Chinese fonts), sizes, bold/italic/underline, colours, paragraph alignment/indentation/spacing/line-spacing, page setup (A4, margins, orientation), the Normal/Heading style defaults and tables — "把 doc 原样转成 docx，但不装 Word、不用 LibreOffice". Commands: - /doc2docx <文件.doc> [-o 输出.docx] - Convert one .doc to .docx - /doc2docx batch <目录> [-o 输出目录] - Batch-convert every .doc in a directory - /doc2docx inspect <文件.doc> - Print structure: FIB, fonts, paragraphs, text preview - /doc2docx check <输出.docx> [--against 原.doc] - Verify the produced .docx and diff text against the original - /转docx <文件> - 中文命令别名（单文件转换） Capabilities: OLE2/CFB parsing (olefile), FIB + CLX piece-table text extraction (UTF-16LE / cp1252), CHPX/PAPX FKP-page decoding into character & paragraph formatting, STSH style-chain resolution, STTBF ffn font-name table (with WPS UTF-16 auto-detection), PlcfSed/SEPX page setup, table reconstruction from fInTable/fTtp marks + TDefTable column widths + TTableBorders, PAGE/NUMPAGES field-code preservation, python-docx + lxml OOXML output with per-run rPr / per-paragraph pPr / per-section sectPr, batch CLI, structural inspect, fidelity check, 23-check self-test with an in-memory synthetic .doc

Lossless DOCX unpack / edit / repack toolkit. Treats a .docx as a ZIP of OOXML parts, edits the raw XML with lxml, and on save copies every UNMODIFIED part byte-for-byte — only the parts you actually edited are re-serialised. Guarantees that opening, editing and saving a document leaves its formatting completely intact. Triggers when: Editing a .docx while preserving format exactly (改文字但不改格式), unpacking a .docx into a folder of raw XML/media/files then repacking it back, swapping an embedded image or font, editing headers/footers/footnotes/endnotes/comments directly, editing table cells without disturbing the table style, doing global find-and-replace across runs while keeping each run's formatting, or any task that requires "open docx → change something → save, and nothing else changes". Commands: - /docx编辑 <文件> - Open a docx in the lossless editor (describe what to edit in natural language) - /docx编辑 unpack <文件> [目录] - Unpack a .docx into a folder of raw parts - /docx编辑 pack <目录> <输出.docx> - Repack a folder back into a .docx - /docx编辑 dump <文件> - Print paragraph & table structure - /docx编辑 replace <文件> <旧> <新> [-o 输出] - Global find-and-replace (preserves formatting) - /docx编辑 set-para <文件> <序号> <文字> [-o 输出] - Rewrite a body paragraph - /docx编辑 set-cell <文件> <表> <行> <列> <文字> [-o 输出] - Rewrite a table cell - /docx编辑 parts <文件> [--bytes] - List every part inside the package - /docx编辑 find <文件> <关键词> - Find paragraphs containing a keyword - /docx编辑 gui [文件] - Launch the PySide6 GUI - /docxedit <file> - English command alias for the editor - /docxedit unpack <file> [dir] - Unpack a .docx into a folder - /docxedit pack <dir> <out.docx> - Repack a folder into a .docx - /docxedit gui [file] - Launch the PySide6 GUI Capabilities: Lossless byte-for-byte preservation of unmodified OOXML parts, lazy lxml parsing with explicit dirty-set serialisation, cross-run text replacement preserving each run's rPr, paragraph rewrite preserving pPr and first-run rPr, table row/column add/delete via deepcopy cloning, in-place header/footer/footnote/endnote/comment editing via raw XML access, embedded image/font swap via set_part_bytes, unpack-to-folder and repack-from-folder with [Content_Types].xml-first ordering, CLI (unpack/pack/dump/find/replace/set-para/set-cell/parts), PySide6 GUI with paragraph/table/parts tabs

Global safety and operational rules for opencode sessions. Auto-loaded with highest priority on every session. Contains critical rules learned from incidents to prevent system crashes and data loss.

Top-tier software architect agent for complex multi-stage project development with strict quality gates and iterative refinement. Triggers when: Designing architecture for a complex multi-stage project, decomposing requirements into modules, or establishing quality gates for development phases. Commands: - /architect design <task> - Full architecture design for a task - /architect phase <n> - Execute specific phase - /architect iterate <module> - Iterate on a specific module - /architect status - Show current architecture status - /architect review - Review and validate architecture Capabilities: Six-phase architecture with quality gates, requirement analysis, architecture design, task decomposition, iterative development, integration and validation, documentation generation

Official document writing assistant for Chinese government documents based on GB/T 9704-2012 standard. Triggers when: Writing Chinese official documents (公文), formatting documents according to national standards, reviewing document compliance, or creating notices, reports, requests, replies, letters, or minutes. Commands: - /gongwen notice <topic> - Write a notice (通知) - /gongwen report <topic> - Write a report (报告) - /gongwen request <topic> - Write a request (请示) - /gongwen reply <topic> - Write a reply (批复) - /gongwen letter <topic> - Write a letter (函) - /gongwen minutes <topic> - Write meeting minutes (纪要) - /gongwen check <document> - Check document compliance - /gongwen format - Show formatting rules Capabilities: 15 document types support, GB/T 9704-2012 compliance, hierarchy numbering rules, font and layout specifications, document structure templates

WeChat Official Account (公众号) full-archive scraper. Given a 公众号 name or biz-id, drive a real browser (Selenium) through 搜一搜 / 文章搜索 to enumerate the account's historical articles, then fetch each article URL, convert the rich HTML to clean Markdown (front-matter + readable body + image references), and save locally with dedup, resumable progress, and polite rate-limiting. Triggers when: User wants to download/archive all articles of a 公众号 by name or ID, convert 公众号 articles to Markdown, batch-save a 公众号's history, or back up a 公众号 locally. Commands: - /公众号 <名称或biz> - 抓取指定公众号全部文章为 Markdown - /公众号 resume <名称或biz> - 从上次断点继续 - /公众号 refresh <名称或biz> - 只抓取新增文章（增量更新） - /公众号 search <关键词> - 在已存档中全文检索 - /公众号 list - 列出已存档的公众号 - /公众号 check - 检查依赖与浏览器环境 - /gongzhonghao <name-or-biz> - English command - /gongzhonghao resume <name-or-biz> - Resume interrupted archive - /gongzhonghao refresh <name-or-biz> - Incremental update - /gongzhonghao search <keyword> - Full-text search across archives Capabilities: Browser-driven article enumeration via 搜一搜/文章搜索, biz-id extraction from article URLs, per-article HTML→Markdown conversion preserving headings/code/images/quotes, image download with local referencing, SHA-256 dedup, resumable progress JSON, incremental update by crawl date, polite delay with jitter, user-data-dir cookie persistence for login state.

手写笔记 OCR 流水线：将纸本扫描图与平板 PDF 导出的手写笔记转成可检索的 Markdown。 触发条件：用户提供一批图片（PNG/JPG）或 PDF，需要 OCR 转文字、做润色/纠错、加摘要索引； 或者需要评估这个流水线本身在人工标注对照集上的 CER/WER/检索召回打分。 命令： - /ocr-notes <目录> - 批量 OCR + LLM 润色 + 生成 MD - /ocr-notes refine <raw_ocr.json> - 仅做润色（OCR 已跑过） - /ocr-notes eval <测试集目录> - 评估打分（需人工标注对照集） - /ocr-notes install - 安装/检查 Python 依赖 - /手写笔记 <目录> - 中文命令入口，等价于 /ocr-notes 能力：PaddleOCR 本地推理（不依赖外部 API）、PDF 自动拆页、LLM 纠错与理顺、 自动生成摘要与关键词索引、CER/WER/关键词召回打分、Markdown 可检索产物。

Complete PR lifecycle management from worktree creation through implementation, commits, PR creation, verification loop, to merge and cleanup. Triggers when: Implementing a feature and submitting a PR, fixing an Issue, or completing merge work after code review. Commands: - /合并 <任务> - Start full PR workflow - /合并 create <任务> - Create PR - /合并 check - Check verification status - /合并 status - View PR status - /pr <task> - English command for PR workflow - /pr create <task> - Create PR (English) - /pr check - Check verification status (English) - /pr status - View PR status (English) Capabilities: Worktree management, feature implementation, atomic commits, PR creation, verification loop (CI + Review + Approval), merge and cleanup

Iterative testing, verification, and improvement supervisor for code quality assurance. Triggers when: User requests iterative testing and improvement, code quality review and assurance is needed, automated testing and feedback loops are required, or multiple rounds of refinement are specified. Commands: - /iterate <n> - Run n iterations of test-improve cycle - /iterate stop - Stop current iteration loop - /iterate resume - Resume current iteration loop - /iterate status - Show current iteration status - /iterate report - Generate iteration report Capabilities: Automated test execution and result analysis, quality metrics tracking across iterations, improvement suggestion generation, convergence detection, and detailed iteration reports

File-operation logging and safe-recovery skill. Intercepts destructive file operations (rm/mv/truncate/overwrite) BEFORE they happen, moves the endangered content into a per-project trash with an append-only JSONL manifest, and restores deleted or overwritten files with SHA-256 integrity verification. Triggers when: User is about to delete/move/overwrite files, asks to recover a deleted file, wants a file-operation audit log, suspects accidental deletion, or needs to guard a destructive batch operation. Commands: - /回收 delete <路径> - 删除前先入回收站（防误删） - /回收 overwrite <路径> - 覆盖前先备份旧内容 - /回收 move <src> <dst> - 移动/重命名并记录 - /回收 list [过滤] - 列出回收站条目 - /回收 log [N] - 查看最近 N 条操作日志 - /回收 restore <id|路径|pattern> - 恢复到原位或指定位置 - /回收 inspect <id> - 查看某条目的元数据与校验和 - /回收 purge [天数] - 清理 N 天前的条目（需确认） - /回收 watch - 列出当前会话受守护的操作 - /回收 guard on|off - 开关守护模式 - /trash delete <path> - English command for safe delete - /trash list [filter] - List trash entries - /trash restore <id|path|pattern> - Restore deleted/overwritten file - /trash log [N] - Show recent N operation log entries - /trash purge [days] - Purge entries older than N days Capabilities: Pre-deletion interception with atomic move-to-trash, append-only JSONL manifest with SHA-256 checksums, stable recovery to original or custom location with integrity verification, overwrite snapshotting, move/rename logging, per-project isolation, configurable age-based purge with confirmation, guard-mode toggle for session-wide destructive-command routing.

AI text humanizer that rewrites AI-generated content into natural human language based on AIGC detection countermeasures. Triggers when: Needing to reduce AIGC detection rates, rewriting AI-generated content into natural language, optimizing text to sound more human-written, or performing humanization polish on AI text. Commands: - /人话 <文本> - Humanize text - /人话 --style formal <文本> - Humanize in formal style - /人话 --style casual <文本> - Humanize in casual style - /humanize <text> - Humanize AI-generated text (English) - /detect <文本> - Detect AIGC features in text - /demo - Generate high-AIGC example and demonstrate processing Capabilities: AIGC detection countermeasures, identifying AI-typical patterns, multi-round iterative AIGC rate reduction, preserving original meaning and logic, supporting both casual and formal styles, bilingual Chinese-English text processing

盘点归纳（Inventory & Report）：从任意杂乱目录盘点某类素材（成果、文献、证书、内容库、名单、资产），先定义分类法，再全目录检索，多源交叉核验，输出结构化汇总（表格/清单/一段话/报告）。泛化于一切"整理一堆东西并交叉核对后给出汇总"的任务。 Triggers when: 需要统计/汇总某主题的内容（年度成果、论文清单、软著台账、获奖名单、学生获奖、文献库、内容收藏、证书清单、资产盘点），需要核对数字/条目是否与多个来源一致，需要把杂乱的档案目录整理成能直接上交的表格与总结。 Commands: - /盘点 <目录> [主题] - 盘点指定目录下的素材并输出汇总表 + 一段话总结 - /盘点 核对 <清单> - 对已有清单做多源交叉核验，输出证据链 - /盘点 报告 - 生成 docx 版盘点报告 - /inventory <dir> [topic] - English alias Capabilities: 分类法设计、按关键词全目录检索（glob+rg）、多源交叉核验（xlsx统计表/官方名单PDF/证书OCR，含WPS异常xlsx的zipfile+XML兜底读取）、xlsx分类汇总表生成（带格式）、一段话总结、docx报告、数字口径以用户确认为准、矛盾点列证据链

Literary ghostwriter skill that mimics the writing style of seven literary masters for creative writing. Extended with long-form novel project management, multi-role writing workflow, and consistency tracking. Triggers when: Writing in a specific author's style, creative writing, script writing, novel writing, long-form fiction planning, chapter continuation, or style imitation exercises. Commands: - /文豪 <作家> <选题> — Single-shot style imitation (7 authors) - /literary <author> <topic> — English single-shot style imitation - /文豪 新建 <项目名> — Create novel project with world-building - /文豪 打开 <项目名> — Open existing project - /文豪 项目 — Show project status - /文豪 规划 — Analyze novel position, plan next moves - /文豪 大纲 [操作] — Generate/modify plot outline - /文豪 章纲 <章节号> — Generate chapter-level outline - /文豪 设定 <世界观元素> — View/modify world-building - /文豪 人物 [角色名] — View/modify character card - /文豪 伏笔 — View foreshadowing tracker - /文豪 续写 — Smart continuation from current position - /文豪 润色 <文本> — Style optimization - /文豪 审稿 [章节] — Editor review - /文豪 去AI味 <文本> — AI-trace reduction - /文豪 回收站 — Recycle bin for deleted content Capabilities: Deep analysis of author's spiritual core, providing concrete methods for authentic imitation, bilingual Chinese-English creation, multiple genre support, long-form novel project management, multi-role writing workflow (planner/character-dev/stylist/editor), consistency tracking, chapter-by-chapter smart continuation, foreshadowing management.

Literary master skill (文宗) — writes in the authentic craft of world-class authors, literary movements, national schools, and historical eras, grounded in deep corpus analysis of ~800 works (public-domain world classics + Nobel Prize collection). Triggers when: Writing fiction/poetry/drama in the style of a specific great author, a literary movement (realism, modernism, magic realism, symbolism, etc.), a national school (Russian/French/English/American/German/Nordic), or an era; building characters the way masters build them; engineering story structure; or needing a powerful general literary craft system beyond surface-level imitation. Commands: - /文宗 <风格目标> <选题> — Write in the craft of an author / movement / era / national school - /文宗 流派 <流派> <选题> — Write in a literary movement (现实主义/自然主义/现代主义/魔幻现实主义…) - /文宗 时代 <年代> <选题> — Write in an era's grammar (古典/中世纪/文艺复兴/19世纪/20世纪…) - /文宗 作家 <作家名> <选题> — Write in one master's signature craft - /文宗 人物 <人物> — Build a master-crafted character card - /文宗 结构 <题材> — Engineer a story skeleton - /文宗 声部 <目标> <文本> — Re-voice a passage in a target craft - /文宗 校核 <文本> <目标> — Verify style fidelity against craft rules - /master <target> <topic> — English equivalent - /master 流派/era/author/character/structure/voice/verify — English subcommands Capabilities: Corpus-grounded craft system, 40+ author signatures, 20+ movements, era grammar, national-school traditions, character-building mechanics, story-skeleton catalog, narrative voice control, free-indirect-speech technique, scene/dialogue craft, anti-AIGC literary writing.

Patent writing assistant for Chinese patents with search, disclosure document writing, workflow guidance, and strategic filing advice. Triggers when: Writing a patent application, creating a patent disclosure document, conducting patent search and analysis, planning patent filing strategy, or seeking guidance on the patent process timeline and requirements. Commands: - /patent search <keywords> - Conduct patent search with analysis framework - /patent disclosure <invention> - Write disclosure document with complete structure - /patent report <invention> - Generate self-search report with blocking risk assessment - /patent workflow - Show patent workflow with timeline and milestones - /patent strategy <invention> - Provide filing strategy advice - /patent check <document> - Check patent document compliance - /patent examples - Show patent writing examples and best practices Capabilities: Patent search platform guidance with analysis framework, disclosure document templates with complete structure, patent three-property analysis with blocking risk assessment, writing tips and best practices with quality checklist, workflow timeline estimation with strategic filing advice, self-search report generation, prior art analysis, novelty and inventiveness assessment, Chinese patent law compliance

Fully autonomous continuous iteration skill that works until time or token budget is exhausted. Triggers when: User requests power iteration, continuous development, uninterrupted work, or sets explicit time or token budgets. Commands: - /强力迭代 <任务> [预算] - Start power iteration - /power-iterate <task> [budget] - Start power iteration (English) Capabilities: Fully autonomous decision-making without user confirmation, automatic requirement understanding and evaluation design, automatic iteration path planning, time/token budget management, periodic progress reporting

Project rebuilding skill that references existing projects to rewrite core features from scratch with team parallel review and persistent execution. Triggers when: Needing to reference an open-source project to rewrite core features, building a similar product from scratch, or deeply refactoring an existing project. Commands: - /rebuild <参考项目> <目标> - Start project rebuild - /rebuild analyze <项目> - Analyze reference project - /rebuild team - Start team parallel review - /rebuild ralph - Start persistent cross-session execution - /rebuild checkpoint - View or restore checkpoints - /rebuild status - View rebuild progress Capabilities: Reference project deep analysis, core feature extraction, multi-role parallel review (architect/developer/tester), cross-session persistent execution, rebuilding core features from scratch

Complete Python multi-project development specification for CLI/GUI tools with unified API, OpenAI function-calling integration, and PyPI publishing. Triggers when: Creating a new Python project with CLI and GUI support, setting up pyproject.toml with README and PyPI publishing, implementing unified API with ToolResult pattern, adding OpenAI function-calling tools integration, or writing standardized tests and documentation. Commands: - /python-project init <name> - Initialize new Python project - /python-project structure - Generate project structure - /python-project api - Implement ToolResult API pattern - /python-project cli - Add CLI with unified flags - /python-project test - Generate test suite - /python-project publish - Setup PyPI publishing Capabilities: Project structure guidance from single-file to package migration, CLI unified flags (-V, -v, -o, --json, -q), Python API with ToolResult dataclass pattern, function-calling with TOOLS + dispatch pattern, bilingual README documentation with auto-screenshots, pytest testing with unified test structure

Scientific figure generator that turns a hand-drawn sketch (or text description) into a publication-quality SVG with guaranteed zero overlaps. Pipeline: read the sketch with a vision LLM → extract structured layout JSON → map to a grid-based SVG template → auto-validate for overlaps/overflow/clipping → iteratively refine until 0 failures. Triggers when: A teacher/colleague asks you to draw a scientific flowchart/diagram, you have a rough sketch (photo/scan/screenshot) and need a clean SVG, natural-language instructions are too vague for direct generation so the vision model must read the sketch first, you need SVG flowcharts with absolutely no text-box overlap or element clipping, or you want to convert a PowerPoint/whiteboard draft into a publication-quality figure. Commands: - /画图 <草图路径> [输出路径] - Full pipeline: read sketch → JSON → SVG → validate → refine - /画图 from-json <layout.json> [输出路径] - Build SVG from a layout.json - /画图 from-text "<描述>" [输出路径] - Build SVG from a text description (no sketch) - /画图 validate <file.svg> - Validate an existing SVG for overlaps/overflow - /画图 template - Show the grid-based SVG template usage - /fig <sketch> [out] - English command alias for the full pipeline - /fig from-json <layout.json> [out] - Build SVG from layout.json - /fig from-text "<description>" [out] - Build from text description - /fig validate <file.svg> - Validate an SVG Capabilities: Vision-LLM sketch reading with structured JSON output, grid-based SVG placement (boxes on grid cells — overlap-free by construction), named anchor-point arrows (top/bottom/left/right/compass), auto text wrapping and font-shrinking (min 9pt), Wong colorblind-safe palette, multi-line CJK+latin labels, automated overlap/overflow/clipping validation, iterative refine loop until 0 failures, feedback-loop arrows (dashed), curved arrows for routing around elements, layout.json reproducibility

Read-only GitHub triage for issues and PRs with evidence-backed analysis and parallel processing. Triggers when: Analyzing GitHub Issues or PRs, generating review reports, or triaging problems. Commands: - /审视 analyze <repo> - Analyze all open issues and PRs in a repo - /审视 issues <repo> - Analyze issues only - /审视 prs <repo> - Analyze PRs only - /审视 <URL> - Analyze a specific issue or PR - /analyze <repo> - English command for repo analysis - /analyze issues <repo> - Analyze issues only (English) - /analyze prs <repo> - Analyze PRs only (English) Capabilities: GitHub data reading, issue classification, evidence collection with permalinks, report generation, parallel background processing

Adversarial dual-scrutiny planning skill. Uses single model to simulate 5 adversarial personas (Pragmatist, IntegrationTester, Researcher, Architect, Creative) in 3-round cross-critique debate. Only defensible insights that survive attacks enter the final plan. Triggers when: Planning complex projects needing maximum rigor, surfacing weak assumptions, blind spots, and over-engineering. Use before implementation to stress-test ideas. Commands: - /双重 <任务> - Start adversarial planning debate - /双重 status - View debate progress - /双重 plan - Generate plan from surviving insights - /adversarial <task> - English command Capabilities: 5-persona single-model adversarial debate (3 rounds), insight survival filtering, executable plan generation from defensible insights only, hard constraint extraction, risk-mitigation pairing, open question surfacing

Central skill registry and management system, auto-loaded on session start for skill discovery and invocation. Triggers when: Listing available skills, loading or activating a specific skill, refreshing the skill registry, or viewing skill help information. Commands: - /skills - List all available skills with descriptions - /skill <name> - Load and activate a specific skill - /skill reload - Refresh skill registry - /skill help <name> - Show detailed help for a skill Capabilities: Auto-discovery of all skills in workspace, unified skill invocation via /skill command, skill registry with metadata and descriptions, version tracking and dependency management

Skill refinement tool that iteratively polishes and improves any skill until it reaches optimal quality. Triggers when: Needing to improve an existing skill, polish skill documentation, optimize skill rules, or fix skill deficiencies. Commands: - /修炼 <技能名> - Refine a specific skill - /refine <skill> - Refine a specific skill (English) Capabilities: Deep analysis of skill deficiencies, improvement plan generation, iterative optimization until standards are met, testing to verify improvements, cultivation report generation

Sleepless autonomous execution skill that never stops until the task is 100% complete, with self-healing and forced delivery guarantees. Triggers when: User needs long-running autonomous work, overnight development tasks, unattended AI execution, or any scenario requiring continuous unattended processing. Commands: - /修仙 <任务> - Start sleepless execution - /修仙 <任务> --budget <级别> - Start with budget level - /修仙 status - Check sleepless status - /修仙 log - View sleepless log - /sleepless <task> - Start sleepless execution (English) - /sleepless <task> --budget <level> - Start with budget level (English) - /sleepless status - Check status (English) Capabilities: Eternal autonomous loop with no voluntary exit, forced delivery mechanism, error self-healing, path switching on failure, cross-session persistence, all task modes support (dev/fix/refactor/test/doc/iterate/integrate/explore)

Comprehensive software development planning and implementation skill for multi-interface Python applications with academic research integration. Triggers when: Creating new Python software with CLI/GUI/Web interfaces, planning software architecture and modules, designing scientific or engineering applications, setting up bilingual documentation and PyPI publishing, or needing academic research-based feature design. Commands: - /planner research <topic> - Conduct domain research - /planner design <project> - Design system architecture - /planner modules - Generate module specifications - /planner docs - Create bilingual documentation - /planner verify - Run verification checklist Capabilities: Pre-development planning and research, multi-interface design (CLI + PySide6 GUI + Flask Web), scientific visualization with pyqtgraph, academic literature-based feature design, sample data and test documentation generation, bilingual README with structured sections, GPLv3 licensing and PyPI publishing setup

Template-preserving form/report filler. Extracts text from mixed reference materials (PDF/OCR/images/Office), corrects misaligned lines and OCR errors, then fills the content into a Word/Excel template while keeping the ORIGINAL fonts, borders, merged cells and layout — never rebuilds from scratch. Triggers when: Filling official forms/reports based on a fixed template (成绩单、试卷质量分析表、成绩分析表、过程性评价档案、考查课报告、各类表格), extracting text from reference materials (PDF/扫描件/图片) then filling into template, when output MUST match the original template format exactly, or when converting .doc/.xls templates and populating them with data. Commands: - /填表 <模板> <数据> - Fill a template with data, preserving format - /填表 convert <文件> - Convert .doc/.xls template to editable .docx/.xlsx - /填表 inspect <模板> - Map a template's tables/cells/fonts before filling - /填表 extract <资料目录> - Extract text from reference materials (PDF/OCR/images) - /填表 check <文档> - Verify filled doc format fidelity & data integrity - /tianbiao <template> <data> - English command for template-preserving fill - /tianbiao convert <file> - Convert legacy template to editable format - /tianbiao inspect <template> - Map template structure & fonts - /tianbiao extract <dir> - Extract text from reference materials - /tianbiao check <doc> - Verify fidelity & data integrity Capabilities: Text extraction from PDF/images/Office (pdfplumber/pymupdf/PaddleOCR/Tesseract), OCR error correction and misaligned-line fixing, in-place cell filling on original templates, .doc/.xls to .docx/.xlsx conversion via LibreOffice, font/border/merge preservation, compact roster tables with repeating headers and non-splitting rows, formula reverse-engineering from source data, format-fidelity verification, bilingual Chinese-English operation

先验证后交付（Verify Before Deliver）：一切任务的交付前强制验证协议。核心主张——任何交付物在汇报给用户之前，必须先真实执行产出、用可复现的方法自测、修复发现的问题、确认无损，最后才汇报。绝不"只描述不执行"，绝不"改了不验证"。 Triggers when: 任何需要交付产物的任务（写代码、改文档、生成表格/图片/PDF、整理文件、填表、论文排版、教材编写、批量处理），特别是用户有"先自己完整测试验证，解决所有问题再汇报""要真实执行不要只描述"期望的场景；也用于交付前的自查清单。 Commands: - /先验证 - 进入交付前验证模式：列出产物、校验、修复、汇报 - /先验证 check <产物> - 对指定产物运行校验清单 - /先验证 报告 - 输出验证结果与修复记录 - /verify <artifact> - English alias Capabilities: 产物可打开性校验（docx/xlsx/pdf/图片/HTML）、内容正确性校验（数字/姓名/奖项/引用一致性）、格式保真校验（字体/合并单元格/页眉页脚）、文件完整性校验（字节级对比、无文件丢失）、跨源一致性校验（统计表/名单/证书多源对齐）、修复-重验循环、交付前汇报规范（先验证→再报结论）

将任意视频网站链接转换为高效率图文学习材料。 Triggers when: 用户提供视频链接并要求下载、截图、逐字稿、幻灯片、总结或浓缩。 Commands: - /视频浓缩 <视频链接> — 完整流程：下载/截图→逐字稿→幻灯片→文章 - /视频截图 <视频链接> — 仅截图并生成幻灯片 - /视频转文字 <视频链接> — 仅提取逐字稿与总结文章 - /视频字幕 <视频链接> — 仅生成带时间戳的字幕文件 Capabilities: 视频下载(合法源)、关键帧提取、语音转文字、基于截图的幻灯片生成、逐字稿+整句字幕+总结文章、多平台支持

视频萃取：将任意视频链接转化为图文学习材料（逐字稿、关键帧、幻灯片、总结文章、网页）。 Triggers when: 用户提供视频链接并要求萃取、浓缩、转录、生成学习材料。 Commands: - /视频萃取 <视频链接> — 完整流程：下载→字幕→文本修正→关键帧→网页+幻灯片+总结 - /VideoExtract <url> — English equivalent Capabilities: 多平台视频下载、音频转文字、关键帧提取、幻灯片生成、总结提炼、网页打包

Auto mouse activity keeper for online course platforms. Prevents video pause caused by mouse-leave detection on platforms like 学习通/雨课堂/智慧树/中国大学MOOC. Triggers when: Need to keep online course videos playing without manual mouse movement, platform pauses video when mouse leaves the player area, or need unattended course playback. Commands: - /网课 - Start default mouse keeper (move mouse every 30s) - /网课 start [interval] - Start with custom interval (seconds) - /网课 stop - Stop mouse keeper - /网课 status - Check if running - /网课 browser - Directly control browser via Selenium (no plugin needed) - /网课 click - Click mode (periodically click play button area) /wangke - English command Capabilities: Cross-platform mouse micro-movement (pyautogui/pynput), direct browser control via Selenium (simulates mouse events on video element, auto-resumes paused video), configurable interval, CLI background execution, anti-detection stealth mode, emergency stop (Ctrl+C or move mouse to corner), supports Windows/macOS/Linux

有据 (Grounded Generation) — 防幻觉的严格转述技能。在知识库检索、网页搜索、 文档整合的每一步生成中，强制执行"闭世界"约束：输出中的每个实体、数字、 因果关系必须能在给定来源中逐字或近逐字找到对应；禁止新增实体、新增数字、 新增逻辑关系。每步生成后运行 grounded 审计，幻觉检出即退回重写。 Triggers when: 检索增强生成 (RAG)、总结多份文档、写文献综述/调研报告、 转述网页内容、向用户汇报搜索结果、整合知识库片段，或任何"内容必须严格 忠于来源"的场景；也用于对已有输出做幻觉审计。 Commands: - /有据 <任务> - 以严格有据模式执行检索/整合/转述任务 - /有据 审计 <输出> [来源目录] - 对已有输出做幻觉审计 - /有据 对照 <输出> <来源> - 逐句对齐输出与来源，标记无据句 - /有据 实验 - 运行基线vs受控对照实验（见 tests/） - /grounded <task> - English command - /grounded audit <output> [sources] - Audit existing output - /grounded experiment - Run baseline vs controlled comparison Capabilities: 闭世界约束（实体/数字/关系三类白名单）、三步一审计 （检索后/生成中/交付前）、逐句引用溯源（[A:L12] 句级标注）、无据句 降级处理（删除或显式标注"来源未提及"）、脚本化实体/数字泄漏检测 （ground_check.py）、基线vs受控 A/B 对照实验框架

Self-controlled autonomous iteration for opencode itself with cross-session memory and safe rollback. Triggers when: Needing opencode self-improvement, continuous skill iteration, or autonomous execution of long-running tasks. Commands: - /自控 <目标> - Start autonomous iteration - /自控 status - View iteration status - /自控 pause - Pause iteration - /自控 resume - Resume iteration - /自控 log - View iteration log - /auto <goal> - English command for autonomous iteration - /auto status - View status (English) - /auto pause - Pause iteration (English) - /auto resume - Resume iteration (English) Capabilities: True autonomous while(true) loop, cross-session context persistence, self-review after each iteration, budget control with time/token management, parallel task handling, decision tree for autonomous choices, rollback mechanism for safe iteration