agent-basic-skill

Use as the workspace-level entry skill for this repository when responses should follow BetterGPT defaults. This skill routes by default to BetterLanguage for language shaping, then applies task-specific enhancements only when needed, such as BetterFrontend for frontend presentation work, BetterVibe for full-stack engineering boundary and maintainability guidance, BetterTrellis for Trellis workflow, Codex soft integration, initialization, spec, task, and hooks-related guidance, and BetterSubagents for subagent delegation, orchestration, wait handling, and takeover guardrails. Use when a single main entry skill should coordinate default loading order and on-demand sub-skill activation.

Use when tasks involve reading, creating, or reviewing PDF files where rendering and layout matter; prefer MinerU API first for parsing/extraction, then use visual checks with Poppler and Python tools such as `reportlab`, `pdfplumber`, and `pypdf` for generation or fallback extraction.

Use when generating, revising, polishing, or reviewing frontend UI, landing pages, dashboards, app screens, marketing pages, or other presentation-layer output that should look like a production-ready product instead of a demo, draft, wireframe, showcase, or developer-facing mockup. Especially use when the request mentions removing AI-like UI, avoiding demo feel, preventing TODO or note text, hiding reasoning or design commentary from users, eliminating placeholder explanations, or making the final interface feel shippable, real, and user-facing.

Use when GPT-5.4 style answers need to be rewritten into concise, natural Chinese with minimal filler, no internet slang or habitual phrasing, and a strict two-part structure that users can scan top-to-bottom at a glance. Especially use when the output should preserve two-part reasoning and solution flow without always printing rigid section titles, allow light line breaks for readability, remove translation-like Chinese, and switch to terminal-friendly rendering when the current surface is CLI or shell rather than an app-style chat surface.

Use when the task involves subagent selection, delegation strategy, parallel execution, wait and timeout handling, fallback takeover, or how to correctly use the project's local agents together with repository skills and MCP tools. Especially use when the user asks how to orchestrate explorer, worker, reviewer, docs-researcher, code-mapper, frontend-developer, browser-debugger, refactoring-specialist, or other local subagents, or when the main agent needs stronger delegation guardrails.

Use when working with Trellis in Codex, especially for soft integration, project initialization, workflow entry, spec and task organization, hooks, generated command structure, or reducing the need to remember Trellis commands manually. Use when the request mentions Trellis init, Codex integration, workflow, spec, task, workspace logs, hooks, update, onboarding, or agent command mapping.

Use when designing, reviewing, or implementing full-stack application structure and engineering boundaries. Especially use when frontend pages are becoming overloaded, presentation code is mixing with backend concerns, backend services are growing into large coupled modules, or the system needs clearer separation between UI, business logic, persistence, and cross-layer responsibilities.

Create an evidence-backed deep research report and final PDF from multimodal sources such as webpages, PDFs, papers, images, videos, transcripts, code, and structured data. Use when the user wants a polished research deliverable with figures, formulas, tables, clickable links, explicit source provenance, and a compileable LaTeX plus rendered PDF rather than a plain-text summary.

Research, outline, write, render, and export end-to-end presentations with Slidev, including speech notes, web-sourced evidence, charts, optional SVG slide assets, and HTML/PDF/PPTX outputs. Use when the user wants a full PPT-style deck from a topic, manuscript, report, or existing slides, especially for technical talks, tutorials, internal reviews, and data-heavy presentations.

Extract, document, compare, and reuse the frontend style of an existing web app or frontend repo. Use when the user wants to read a frontend codebase, distill its visual and implementation style into a reusable style pack, then apply or enforce that style in new pages, components, or products.

**REQUIRED** - Activate first for any VCS work in jj-enabled repos. Trigger when `.jj/` exists, repo docs require `jj`, or the user explicitly asks for jj commands. Work JJ-first, prefer dedicated workspaces for parallel tasks, avoid interactive flows, and create bookmarks late.

Use when the user wants Codex to launch one or many `codex exec` workers from prompts or prompt files, keep JSONL/stderr/final-message logs on disk, monitor live run status from files, and collect batch results. Best for orchestration, automation, and parallel case execution. Focus on CLI flags, config overrides, environment variables, run directories, and status files; do not spend time on Codex DB internals or memory internals.

用于调试 Tauri 应用的 MCP 工具集，提供截图、DOM 检查、元素交互、IPC 监控等功能。Use when working with Tauri applications for: (1) Debugging UI issues with screenshots and DOM inspection, (2) Automating UI interactions like clicks and input, (3) Monitoring IPC communication between frontend and backend, (4) Managing Tauri windows and sessions, (5) Reading console and system logs.

基于 browser-use CLI 的浏览器自动化调试工具集，提供网页控制、元素交互、截图、Cookie 管理、云端浏览器等功能。Use when working with web applications for: (1) Browser automation and UI testing, (2) Taking screenshots and inspecting page elements, (3) Filling forms and simulating user interactions, (4) Managing cookies and sessions, (5) Using cloud-based browsers for remote debugging, (6) Creating tunnels for local development servers.

Investigate public AI access resources across forums, communities, directories, search engines, and the open web. Use when Codex needs to search one platform such as Linux.do or a broader web scope for public welfare stations, relay sites, signup flows, model-support claims, availability reports, public base URLs, or status signals, and return a reusable evidence ledger for later validation. Do not use this skill to harvest or republish unauthorized secrets; record only public URLs and redacted or user-authorized secret references.

Probe user-provided or user-authorized LLM relay or gateway endpoints, classify whether they actually support OpenAI chat, OpenAI responses/Codex, or Anthropic messages, and optionally wire confirmed working relays into the local CCH/check-cx stack. Use when the user gives a base URL plus an authorized credential and asks to test it, list working models, explain relay failures, or add the relay locally.

Use when the user wants to turn a YouTube or Bilibili lecture, tutorial, or technical talk into a structured Chinese LaTeX/PDF note that preserves the video's real teaching content, keeps the official cover on the front page, selects figures with subtitle-aligned high-recall frame search, ends with a final synthesis chapter, and routes through the unified video-note wrapper plus an external runtime repo.

Use when the user wants to turn a YouTube or Bilibili lecture, tutorial, or technical talk into a structured Chinese LaTeX/PDF note that preserves the video's real teaching content, scales coverage roughly with subtitle and time emphasis, selects figures through writing-driven evidence or explanation needs, ends with a final synthesis chapter, and routes through the unified video-note wrapper plus an external runtime repo.

Use when the user wants to turn a YouTube or Bilibili lecture, tutorial, or technical talk into a structured Chinese LaTeX/PDF note that preserves the video's real teaching content, enforces subtitle-char length guardrails, routes figure strategy through carrier-family/support-profile/recall-budget, records visual obligations explicitly, and closes with a real post-build revision loop through the unified video-note wrapper plus an external runtime repo.

Use when the user wants to maintain daily Bilibili creator digests through the external PulseDeck repository and a local Obsidian bilibili vault.

Use when the user needs cache-first or live-search access to Shuiyuan topics through the external shuiyuan_exporter repository.

Use when the user wants to turn one or more Bilibili video URLs into Markdown notes with the external BiliNote repository, without starting the frontend/backend web services.

Use when the user asks to create, scaffold, or edit Jupyter notebooks (`.ipynb`) for experiments, explorations, or tutorials; prefer the checked-in templates and run the helper script `new_notebook.py` to generate a clean starting notebook.

Use when the user wants to create or update a Chinese deep paper note by driving the external paperflow repository and a local research-notes workspace.

Use when the user asks to download A-share or Hong Kong listed company financial report PDFs, including 年报/中报/一季报/三季报, annual report, interim report, or when a stock-analysis workflow needs the latest full-year report PDF.

Create stunning, animation-rich HTML presentations from scratch or by converting PowerPoint files. Use when the user wants to build a presentation, convert a PPT/PPTX to web, or create slides for a talk/pitch. Helps non-designers discover their aesthetic through visual exploration rather than abstract choices.