InsightEngine

Turn stories, folktales, or existing prose into a comic production package. Use when the user asks to chuyển thể truyện tranh, create a manga/anime-style adaptation, build character prototypes, write a character bible, plan comic pages, storyboard scenes, or keep visual continuity across a comic series. Also use when the user wants the workflow to go from story analysis → character design → page plan → visual reference sheet for reuse. Always use this skill when the user says: "chuyển thể truyện tranh", "vẽ thành manga", "tạo character sheet", "nguyên mẫu nhân vật", "thiết kế nhân vật truyện tranh", "tạo bảng tham chiếu nhân vật", "storyboard truyện tranh", "page plan", "kế hoạch page", "character bible", "giữ nhất quán nhân vật", "vẽ nhân vật", "comic adaptation", "manga adaptation", "anime character design" — even if they don't say "comic-adaptation".

Generate charts, AI illustration images, and train character LoRA for consistent generation. Chart mode (matplotlib): bar, line, pie, radar, scatter — from Excel/CSV/inline data, output PNG. Image generation mode (AI): characters (nhân vật), backgrounds, landscapes, and slide backgrounds/frames — auto-detects GPU (CUDA/MPS/CPU), no special hardware required. LoRA training mode: train a character LoRA from reference images using SD 1.5 + IP-Adapter multi-angle pipeline → dataset preparation → LoRA fine-tuning → inference test. Produces a small .safetensors adapter (~6 MB) that maintains character identity across any pose/scene. Always use this skill when the user wants any visual asset: data charts, illustration images, background images for slides, characters or scenes for documents, slide frames, or anything visual to make content look more professional — even casual requests like "tạo hình nền", "vẽ nhân vật", "cần ảnh minh họa", "làm slide đẹp hơn", "tạo background", "vẽ cái gì đó", "tạo biểu đồ", "vẽ char

E2E testing via Chrome DevTools MCP — test authenticated flows (Google OAuth, SSO, sessions) by connecting to the user's real Chrome browser with existing login state. Also provides DevTools-level debugging: performance traces, Lighthouse audits, network inspection, console monitoring, and memory profiling — capabilities Playwright cannot offer. Use this skill when: the user needs E2E tests through their real browser, testing requires auth/cookies/sessions, testing Google OAuth or SSO flows, debugging live pages with DevTools, running Lighthouse or performance analysis, or sites block bot-driven browsers. Triggers: "test qua Chrome", "kiểm tra e2e", "test trên trình duyệt thật", "test auth flow", "test Google login", "chạy e2e qua MCP", "test SSO flow", "E2E test with auth", "check my app in Chrome", "run lighthouse", "kiểm tra performance", "debug trên Chrome", "chạy Lighthouse", "test trang web", "check network requests", "test live site". Prefer over webapp-testing when auth/cookies matter or DevTools feat

Search the internet for information, discover platforms, and collect structured data. Primary search: Playwright browser → DuckDuckGo/Google HTML scraping. Fallback: httpx zero-auth → DuckDuckGo HTML. Last resort: vscode-websearchforcopilot_webSearch. 3-tier URL fetching for results: fetch_webpage → httpx → Playwright stealth mode. Three modes: standard search (single topic), deep research (multi-dimension exhaustive), and data collection (structured items with direct URLs from specific platforms). Auto-reviews search quality and expands queries if content is insufficient. Always use this skill when the user wants to find information online, research a topic, or collect structured data from platforms — "tìm kiếm về X", "search Google", "tìm hiểu", "danh sách việc làm", "so sánh các nền tảng", or when the pipeline needs web content. Do NOT use for reading local files or fetching explicit user-provided URLs → use gather.

Create professional Excel (.xlsx) files with data, formulas, and formatting. Uses openpyxl for formatting/formulas, pandas for data operations. Runs scripts/recalc.py after generation to force formula recalculation. Always use this skill when the user needs a spreadsheet, table with calculations, or data export — even casual requests like "làm cái bảng tính", "tạo file excel", "xuất ra bảng", "tính toán và lưu thành file", or "cho tôi bảng so sánh" where tabular data with formulas fits, even without saying "gen-excel" or ".xlsx".

Create professional static HTML pages OR reveal.js browser presentations from synthesized content. Two modes: "page" (static page with inline CSS) and "presentation" (reveal.js slides). 8 styles with light/dark variants, transitions, fragment animations, code syntax highlighting. Note: presentation mode requires internet connection for reveal.js CDN. Always use this skill when the user wants a webpage, HTML report, or browser-based presentation — even casual requests like "tạo trang web", "làm slide trình chiếu trong trình duyệt", "xuất HTML", or "tạo cái gì đó mở được bằng Chrome" — even without saying "gen-html".

Create professional PDF documents from synthesized content, with Vietnamese font support. Uses reportlab Platypus for complex multi-section layouts, Canvas for simple single-page PDFs. Always use this skill when the user wants a PDF output — even casual requests like "xuất PDF", "tạo file pdf", "lưu thành PDF để in", "tôi cần file pdf", or "cho tôi file không chỉnh sửa được" (read-only document implies PDF) — even without saying "gen-pdf" or ".pdf".

Create professional PowerPoint (.pptx) presentations from synthesized content. Two engines: Quick mode (pptxgenjs, 10 templates) for pipeline tasks, and Pro mode (ppt-master SVG→PPTX) for consulting-grade output with native DrawingML shapes, 20+ layout templates, 50+ chart templates, and 6700+ icons. Colors must be hex without # prefix (pptxgenjs format requirement). Always use this skill when the user wants a PowerPoint, deck, or .pptx file — even casual requests like "làm bài thuyết trình", "tạo slide", "tôi cần deck để present", "xuất ra PowerPoint", "cho tôi file pptx", "tạo bài slide chuyên nghiệp" — even without saying "gen-slide". Use Pro mode when user says "slide chuyên nghiệp", "consulting-grade", "slide đẹp thật sự", "dùng ppt-master", or has source docs (PDF/DOCX/URL) to convert.

Create professional Word (.docx) documents with 3 template styles: corporate, academic, minimal. Uses python-docx with A4 page setup. Supports tables, headings, TOC, and custom formatting. Always use this skill when the user wants any Word or .docx output — even casual requests like "lưu vào file word", "tạo tài liệu", "làm cái báo cáo word", "export text ra file", or "cho tôi file để gửi sếp" where a Word document is clearly the right format, even without saying "gen-word" or ".docx".

Synthesize and merge multi-source content into coherent, expert-level documents. Default mode is COMPREHENSIVE — rich, analytical content with specific data, examples, and insights. Auto-reviews its own output: checks depth, specificity, and analytical quality per section. If any section is too thin or generic, automatically rewrites it before delivery. 4 modes: standard (brief), comprehensive (default — rich expert-level), translation (Vietnamese↔English), summary. Identifies overlapping content, resolves conflicts, proposes outline before writing. Always use this skill when the user has multiple pieces of content to combine, wants to translate, wants to expand brief notes into a full document, or says things like "gộp lại", "tổng hợp nội dung", "dịch sang tiếng Anh/Việt", "viết lại đầy đủ hơn", "biên soạn", "synthesize", "merge content" — even if they don't say "compose" explicitly.

Create professional visual designs programmatically — cover pages, posters, certificates, infographic layouts, title pages, invitations, and artistic compositions. Uses reportlab Canvas and Pillow to draw shapes, typography, and patterns directly in code, with 80+ bundled fonts. Output: PNG or PDF. Always use this skill when the user wants a designed visual composition with typography, layout, and artistic intent — "tạo poster", "thiết kế bìa", "làm certificate", "tạo thiệp", "thiết kế cover page", "tạo bằng khen", "tạo banner", "design a poster", "make a cover", "tạo infographic", "tạo hình bìa báo cáo" — even without saying "design". Do NOT use for data charts (use gen-image) or AI-generated illustrations (use gen-image image mode).

Read content from local files and explicit URLs provided by the user. Uses markitdown as primary reader with format-specific fallbacks for garbled output. 3-tier URL fetching: fetch_webpage → httpx → Playwright stealth mode (for bot-protected sites). Auto-reviews gathered content quality and retries with fallback readers if insufficient. Always use this skill when the user provides files to read or specific URLs to fetch — "đọc file này", "lấy nội dung từ URL này", or when a file path or URL is dropped into chat. Do NOT use for web search, topic research, or platform discovery → use search skill.

Setup and install all InsightEngine dependencies. Runs check_deps.py first, installs only missing packages. Creates utility scripts (recalc.py, save_state.py) if not present. Always use this skill when the user gets an import error, ModuleNotFoundError, or says the tool isn't working — even casual requests like "bị lỗi import", "thư viện chưa cài", "lần đầu chạy", "không chạy được", "cài lại đi", or "setup môi trường" — even without saying "setup".

Content synthesis skill for InsightEngine — gathers, merges, and structures multi-source content into coherent documents. Called by the orchestrator orchestrator agent or directly via synthesize. Handles the core content pipeline: gather → compose → gen-[format] with auto quality review. Default content depth is COMPREHENSIVE (expert-level, rich content). Supports session resume via save_state.py and chained outputs (e.g., Excel data → chart → slide). NOTE: Orchestration (intent classification, routing) is handled by orchestrator agent. This skill focuses purely on content synthesis workflows.

Intelligence-driven audit: Copilot READS output content, OPENS URLs to verify, COMPARES data against actual web pages, and REASONS about quality — not just script-based rule checks. The key difference: this skill makes Copilot act as a human reviewer who actually clicks links, reads pages, and judges whether the output is truthful and complete. Works as synthesize Step 4.7 (automatic) or standalone when user says "kiểm tra", "audit", "check xem đúng chưa", "sai ở đâu", "thiếu gì", "verify output".

Session retrospective and continuous improvement skill for InsightEngine. Analyzes the entire work session — user's original request, intermediate steps, final output, quality gaps — and produces actionable improvement recommendations for the pipeline, individual skills, and overall process. Can also create new skills or modify existing ones to address systemic issues. Always use this skill when: the user says "cải tiến", "retrospective", "phân tích session", "tại sao kết quả không tốt", "cải thiện quy trình", "improve pipeline", "analyze what went wrong", "session review", "process improvement", "nâng cấp skill", "tại sao output kém", "lesson learned", or when the user is frustrated with output quality and wants systemic fixes rather than just a redo. Also use when the user explicitly asks to improve a specific skill based on real usage experience, or when a pattern of failures is noticed across multiple runs. Do NOT use for one-off output fixes (use verify for that) or for creating skills from scratch witho

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Advanced skill creation and refinement pipeline. Wraps skill-creator with an automatic multi-criteria review loop that grades skills on quality, usefulness, alternatives analysis, and implementation — then iteratively improves until all criteria reach grade A, or stops after 10 loops with no further improvement. Use this skill whenever the user wants to create a high-quality, production-grade skill with rigorous self-review, or wants to upgrade/audit an existing skill to professional standards. Also triggers when the user says "tạo skill nâng cao", "forge a skill", "create and review skill", "auto-improve skill", "skill chất lượng cao", "nâng cấp skill", "review skill", "audit skill quality", or wants a skill that's been validated through multiple rounds of critical analysis rather than just a first draft.