메뉴
Control the VTuber avatar — start and stop the renderer, virtual camera, and virtual mic; speak through the avatar with lip sync; respond to voice conversation when hey-bot is active.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Authoring AI instruction surfaces - SKILL.md, agent definitions, CLAUDE.md policies, and subagent briefs. Use when writing or editing any file that instructs an AI.
Interact with a live webpage inside a browser window — fill forms, click buttons, navigate authenticated apps, automate Electron apps, test web UI, capture browser-window screenshots for visual verification of web pages. Do NOT use for fetching or reading web content programmatically (use curl or MCP fetch). Do NOT use for desktop screenshots, system media/audio control, MPRIS players, or non-browser GUI — those belong to the desktop skill.
Drive herdr, the terminal workspace manager, and orchestrate Claude Code agents in it: spawn, prompt, read, wait on, or take over a background Claude. Also scripts herdr workspaces, tabs, and panes.
Code review, compliance checking, documentation standards, instruction-file audits, and output compression. Use when reviewing code, checking conventions, writing docs, auditing skill or CLAUDE.md edits, or when output needs to be more concise.
Manage the chise media stack: create and manage Jellyfin friend accounts, and check Jellyseerr request and download status. Use for friends, media requests, or Jellyfin users.
Log into and scrape data from Lucas's B3 Área do Investidor (investidor.b3.com.br) via the PinchTab browser-automation CLI plus raw CDP XHR capture. Use when syncing portfolio positions, trades, dividends, or any other read-only data from B3 — for example when Golden runs b3-sync, or when verifying brokerage state. Documents the working flow, the SPA quirks that broke prior attempts, the API endpoint catalog, and the institution-CNPJ filter trick.
| name | avatar |
| description | Control the VTuber avatar — start and stop the renderer, virtual camera, and virtual mic; speak through the avatar with lip sync; respond to voice conversation when hey-bot is active. |
<voice_conversation_mode> When avatar is active with hey-bot daemon, set up a cron job polling transcription logs every 30s. On events: read tail of the transcription log, filter out noise (nonsensical text, self-TTS re-transcription, entries older than 60s), respond to genuine human speech via avatar-speak.sh. Keep responses concise. Respond via avatar, not Telegram. </voice_conversation_mode>
Virtual camera not in Meet: restart Meet (Chrome enumerates devices at join time, not dynamically). No audio in Meet: check pactl for AvatarMic sink, use "mic" output mode. Speak hangs: control server must be running (check health endpoint). Renderer won't start: npm install in the renderer directory.