dotfiles

Kimi WebBridge lets AI control the user's real browser — navigate, click, type, read, screenshot, and interact with any website using the user's actual login sessions. Use this skill whenever the user wants to interact with websites, automate browser tasks, scrape web content, or perform any action requiring a real browser. Also use when the user mentions "browser", "webpage", "open URL", "screenshot", or asks to read/interact with any website. Use even for simple-sounding browser requests — the daemon handles all complexity.

Record demo GIFs/MP4s of agent TUI sessions (Claude Code, Codex CLI, other agent CLIs), synchronized via a Stop-hook sentinel so the recording knows exactly when each response finishes. Two modes - scripted (VHS tape, repeatable) and live/adaptive (tmux + asciinema, where the outer Claude reads each inner response and decides the next prompt). Use when the user wants to record, capture, or make a demo/GIF/video of a Claude Code or Codex session, create a .tape file, drive a nested agent session interactively, or asks about VHS/asciinema recording of an agent CLI.

Put a website behind a Cloudflare Access (Zero Trust) login gate, or remove one, entirely from the CLI — no dashboard GUI. Use when the user wants to password/email-protect a hostname, gate a Cloudflare Pages or Workers site, restrict a site to specific emails, set up Zero Trust Access, or asks about "cf-gate". Manages Access applications and allow-email policies via the Cloudflare API using a token in the skill's .env. Note: wrangler does NOT manage Access — this uses the Cloudflare REST API directly.

Defer execution of a slash-command or prompt until a timer elapses. Starts a background polling timer that prints "TIMER DONE" on stdout, then executes the deferred prompt.

Generate an image via Codex CLI (OpenAI gpt-image-1) and save to a local path. Use when user asks to create/draw/generate an image AND save it (e.g. "draw X, save to images/y.png", "用 codex 生圖"). Requires `codex login` + `$OPENAI_API_KEY`.

DeepSeek client via the private chat.deepseek.com API (browser-exported userToken), not the official platform.deepseek.com API. Use for personal webgui automation, `x-ds-pow-response` / DeepSeekHashV1 POW solving, or SSE THINK/TOOL_SEARCH stream parsing.

End-to-end data-first research pipeline — find an underutilised public dataset, pick the publishable claim, target a journal, ship LaTeX manuscript + tagged GitHub release. Use for computational-bio / bioinformatics / clinical-genomics on TCGA, GEO, GDC, cBioPortal, or when user says "end-to-end paper" / "what's publishable here".

Todoist CRUD via API v1 — tasks (due + deadline), projects, sections, labels, comments, reminders. Use when user mentions Todoist, their todo list, scheduling/completing a task, "加到待辦", "今天的 Todoist".

Author and iterate on `.argdown` argument-map files. Use when the user asks to "create an argdown", structure a debate / argument / pro-con map, lint a `.argdown` file, or export an argument map to SVG/DOT/HTML/JSON via the `argdown` CLI. Workflow — draft, then run `argdown` on the file to lint with line/col errors, fix, then `argdown map -f svg` to render.

Create new Claude Code skills with proper structure. Use when adding a skill or slash command.

Build LLM apps, RAG systems, and prompt pipelines. Use for AI-powered features.

Design system architecture and component interfaces. Use for architectural decisions.

Git checkpoint management. Use when starting risky changes or needing recovery points.

Analyze Claude Code context efficiency. Use when evaluating session token usage.

Debug errors and unexpected behavior with log analysis. Use for issues and error messages.

Create strategic plan and dev docs. Use when starting large multi-session tasks.

Improve developer experience and tooling. Use when setting up projects or reducing friction.

Scan environment for dev readiness — tools, runtimes, permissions. Use in new containers.

Manage Tailscale Funnel routes. Use when configuring tunnels or external HTTPS access.

Enable, disable, and manage Claude Code hooks. Use for hook debugging or configuration.

Scan and replace AI-flavored vocabulary in English text. Use when polishing writing.

Read Mail.app inboxes and create Reminders for actionable items. Use for email triage.

Generate semantic codebase symbol map. Use when navigating unfamiliar code or before refactoring.

Profile and optimize bottlenecks. Use for performance issues or caching strategies.

Load codebase context by analyzing structure and README. Use when starting a new session.

Optimize prompts for LLMs. Use when crafting system prompts or improving agent performance.

Publish a skill folder as a .skill package to GitHub. Use when distributing skills.

Refactor code for quality and maintainability. Use for cleanup and tech debt reduction.

Initialize GitHub repo with CITATION.cff, LICENSE, README badges. Use for new repos.

Three-phase RPI workflow (Research, Plan, Implement). Use for non-trivial code changes.

Session retrospective report. Use when reviewing or summarizing a work session.

Unified SuperClaude workflow with subcommands. Use for structured development workflows.

Expert web research with advanced techniques. Use for deep research or fact verification.

Security audits and secure coding. Use for vulnerability detection or auth implementation.

Create unit, integration, and e2e test suites. Use when setting up or improving tests.

Update dev docs before compaction. Use before ending sessions or switching tasks.

Run Vale prose linter and auto-fix issues. Use when linting or proofreading markdown/text.

Deploy Docker services on VM with Traefik and Homepage. Use for VM container deploys.

Run a comprehensive code review