Creator HQ 대시보드 스킬 (creator-hq) — 주간 운영 현황 확인

creator-hq는 반복 업무를 더 안정적으로 처리하기 위한 AI 에이전트 스킬입니다.

Build and serve the Creator HQ dashboard - a shared home screen for a YouTube creator (and partner) showing a weekly checkup (did we create/release our target videos this week?), published videos with live stats auto-pulled from YouTube, channel KPIs, and a hand-edited production pipeline (idea to ready). Self-contained HTML, servable to an iPad over the LAN from a Raspberry Pi. Triggers on: creator hq, creator dashboard, content dashboard, youtube dashboard, show my channel dashboard, are we on track this week, did we release our videos, content home screen, studio dashboard.

사용 방법

Codex나 호환되는 AI 에이전트의 스킬 폴더에 아래 SKILL.md 내용을 저장한 뒤, 설명에 포함된 트리거 문장으로 호출하세요.

사용 예:
creator-hq 스킬 사용해줘.
이 작업은 creator-hq 기준으로 진행해줘.

원본 위치

.codex/skills/gocut-creator-hq/SKILL.md

SKILL.md

---
name: creator-hq
description: Build and serve the Creator HQ dashboard - a shared home screen for a YouTube creator (and partner) showing a weekly checkup (did we create/release our target videos this week?), published videos with live stats auto-pulled from YouTube, channel KPIs, and a hand-edited production pipeline (idea to ready). Self-contained HTML, servable to an iPad over the LAN from a Raspberry Pi. Triggers on: creator hq, creator dashboard, content dashboard, youtube dashboard, show my channel dashboard, are we on track this week, did we release our videos, content home screen, studio dashboard.
argument-hint: (no args to rebuild) | "serve" | "pipeline" (help me edit the board)
allowed-tools: Bash, Read, Write, Edit
user-invocable: true
---

Build the **Creator HQ dashboard**: one self-contained `dashboard.html` that answers "are we on track this week?" at a glance, shows what's published and how it's doing, and tracks how many videos are in the pipeline. Designed as a shared home screen for a creator and a partner, viewable on a wall/desk iPad served from a Raspberry Pi over the local wifi.

## How it works (no app, no API key, no secrets)

Three small pieces, all in this skill folder:
- **`fetch_stats.py`** - pulls channel meta + recent uploads via `yt-dlp` (no API key), downloads thumbnails, writes a snapshot to `~/creator-hq/data/youtube.json`.
- **`dashboard.py`** - pure renderer. Reads the config + the snapshot + the hand-edited pipeline, computes the weekly checkup, and bakes a fully self-contained `~/creator-hq/dashboard.html` (thumbnails inlined as data URIs - works offline, passes the Artifact CSP).
- **`serve.py`** - serves `~/creator-hq` to the LAN so the iPad can open it; can auto-rebuild on an interval.

Personal data lives **outside this repo** in `~/creator-hq/` (config, pipeline, snapshot, built HTML). Nothing sensitive is committed - the channel handle and goals live in `~/creator-hq/config.json`, which is created from `config.example.json`.

## First-time setup

1. `mkdir -p ~/creator-hq`
2. Copy `config.example.json` to `~/creator-hq/config.json` and set the creator `handle` (e.g. `@YourChannel`) and weekly `goals`. Add a second creator object to `creators` later to cover a partner's channel.
3. Copy `pipeline.example.json` to `~/creator-hq/pipeline.json` and replace the example rows with real ones (or start empty: `{"videos": []}`).

## Rebuild flow

```bash
python3 ~/.claude/skills/creator-hq/fetch_stats.py     # pull latest YouTube stats
python3 ~/.claude/skills/creator-hq/dashboard.py       # render dashboard.html
open ~/creator-hq/dashboard.html                       # or serve it (below)
```

`fetch_stats.py` makes ~3 yt-dlp calls per creator (channel meta, total count, recent N with full metadata) and caches thumbnails, so a rebuild is quick after the first run.

## Serving it to the iPad (Raspberry Pi kiosk)

```bash
python3 ~/.claude/skills/creator-hq/serve.py --rebuild 30
```

Prints a `http://<pi-ip>:8080/` URL - open that in the iPad's browser (or your existing kiosk). `--rebuild 30` re-pulls YouTube + re-renders every 30 minutes; the served page also self-refreshes every 15 min via a meta tag, so the display stays current on its own. To run it unattended, add `serve.py` to the Pi's existing kiosk/startup (systemd service or `@reboot` cron).

## What's on the page (Command Center layout)

- **Weekly checkup band** - three big cards: *videos created* this week (from the pipeline's `filmed` dates) vs goal, *videos released* this week (from real YouTube upload dates) vs goal, and *ready to publish* (pipeline items in the `ready` stage). Green = goal met, amber = partial, red = none yet.
- **KPI row** - Subscribers, Total videos, Avg views (last N uploads) with a sparkline, and Top recent video.
- **Recent uploads** - thumbnail cards with live view counts, likes, publish date, and a duration/SHORT badge. Each links to the video.
- **Pipeline bar** - idea -> scripted -> filmed -> editing -> ready with a count per stage; the ready stage is highlighted green.

## The pipeline file

`~/creator-hq/pipeline.json` is hand-edited (by you or your partner). Each row:
```json
{ "title": "...", "stage": "editing", "created": "2026-07-07", "filmed": "2026-07-09", "notes": "" }
```
- `stage`: one of `idea`, `scripted`, `filmed`, `editing`, `ready`.
- `filmed` date drives the "created this week" checkup (a video "exists" once filmed). Falls back to `created` if `filmed` is null.
- Delete a row once the video is published - it then shows up automatically from YouTube.

When the user says "we filmed X" / "move X to editing" / "X is ready", update `pipeline.json` for them, then rebuild.

## Config / targets

`~/creator-hq/config.json`:
- `creators`: list of `{ name, handle }` (handle like `@YourChannel`). One for now; add more to track a partner.
- `goals.create_per_week` / `goals.release_per_week`: the weekly checkup targets.
- `week_start`: `monday` (default) or `sunday`.
- `recent_count`: how many recent uploads to pull (default 15).

## Notes
- Everything auto-pulled is real (yt-dlp), so no manual stat entry. The only hand-maintained file is the pipeline.
- The built HTML is fully self-contained (offline-safe, Artifact-safe). To preview it on the iPad instantly via claude.ai, publish `~/creator-hq/dashboard.artifact.html` as an Artifact.
- Multi-creator (a partner's channel side by side) is a future extension - the data model is already a list; v1 renders the first creator.