dashboard

name	dashboard
description	Start the local viralman dashboard at http://localhost:8765 — three pages (Twitter / Reddit / Gitmail) with header tab switching, a free-text "what to write" prompt that drives AI generation per platform, unified top-right login, and a language switcher.
level	2

Dashboard skill

The dashboard is three separate pages sharing one project state. Each page runs the same flow: enter project info → say what to write → AI generates a platform-tuned draft → tweak targets → send.

This skill exists so phrases like "open the dashboard", "viralman 띄워줘", "run viralman" route to a deterministic flow.

Trigger phrases

Auto-trigger on:

• /dashboard
• "open the dashboard", "start the dashboard", "viralman dashboard"
• "viralman 켜줘", "viralman 띄워줘", "대시보드 열어줘"
• bare viralman in chat that clearly means "start the app", not the project

한국어:

• "대시보드 띄워줘", "대시보드 켜줘", "대시보드 실행"
• "viralman 웹 UI 열어줘", "viralman 화면 띄워줘"
• "로컬 서버 켜줘", "viralman 서버 실행"

English:

• "open the dashboard", "launch viralman dashboard", "fire up the dashboard"
• "start the viralman web UI", "give me the viralman page"
• "boot viralman locally"

中文:

• "打开 viralman 面板", "启动 viralman 仪表盘"

日本語:

• "viralman の ダッシュボード を 開いて", "ダッシュボード を 起動"

If the user typed /dashboard, follow commands/dashboard.md for argument parsing.

What the pages are

URL	Platform	What you can do
/twitter	X (Twitter)	tweet body + char count + sniffer flags + hashtag chips, post via API or compose-URL fallback
/reddit	Reddit	title + body, subreddit chips, scan recent threads to comment on, post via PRAW
/gitmail	Email outreach	email template (uses {{login}} and {{starred_repo}}), recipient slider, live progress + per-recipient preview

Every page has the same top blocks:

1. project — name, GitHub URL, one-line pitch, multi-line description. Persists in localStorage so it carries across page tabs.
2. what to write — a free-text textarea where the user describes the angle, tone, what to emphasize. Replaces the old "mode" dropdown — the LLM gets the user's intent verbatim.
3. provider — auto / claude / openai / gemini.
4. generate — calls /api/generate, returns a platform-specific draft.

Header (shared across all 3 pages):

• Tabs on the left: Twitter | Reddit | Gitmail (active tab highlighted).
• Language picker mid-right: EN / 한 / 中 / 日, persists in localStorage, auto-detects navigator.language on first visit.
• Connect 0/4 ▾ dropdown on the right — the single login surface. OAuth (X / Reddit / LinkedIn) plus manual tokens (Gitmail SMTP + LLM key) all live there.

Step 0 — environment check (auto-recovery before anything else)

Before launching, verify viralman is installed and importable. Sequence:

1. Resolve the viralman binary using the priority order in commands/dashboard.md step 1.
2. If nothing is found, do NOT attempt to clone or pip-install yourself — invoke /viralman-setup instead. Its Step 0 detects the missing environment and runs the bootstrap (clone, venv, flask, shim) before asking about credentials. Idempotent.
3. After bootstrap (or if already present), confirm the binary's --help works without a ModuleNotFoundError. If flask is missing, run <venv>/bin/pip install flask and retry once. After that retry fails, surface the error.

Why two skills:

• /dashboard is "make the UI run for me, now". It must not become a heavy installer.
• /viralman-setup is the single entry point for "get viralman working" — its Step 0 owns clone, venv, and shim; later steps cover credentials.
• The dashboard skill invokes the setup skill on cache miss. The user shouldn't need to know the difference.

Step 1 — Start the server

viralman                            # if installed
./bin/viralman                      # from repo root
./scripts/dashboard.py              # equivalent

Default flags: --host 127.0.0.1 --port 8765, browser auto-opens to /twitter. Pass --no-browser if no auto-open is wanted.

If Flask is missing, the entry script prints a precise install hint — relay it verbatim. Do NOT pip-install on the user's behalf without consent.

Step 2 — Tell the user what's up

viralman dashboard → http://localhost:8765
  twitter:  http://localhost:8765/twitter
  reddit:   http://localhost:8765/reddit
  gitmail:  http://localhost:8765/gitmail
Ctrl-C to stop.

Say the language switcher and Connect dropdown are both top-right.

Step 3 — Hands off

The user drives from the browser. Don't shadow-post on their behalf, and don't call the post-API endpoints from inside chat — those are gated behind the dashboard's explicit "post" / "send" buttons for a reason.

Common issues

• Port already in use: viralman --port 8766 (or whatever).
• Connect counter stuck at 0/4: no creds saved yet. Point the user at the per-row "tokens" / "setup" buttons in the dropdown, or at the /viralman-login-{reddit,twitter,linkedin,gitmail} skills.
• OAuth login button does nothing: the platform's CLIENT_ID / CLIENT_SECRET aren't saved. Each Connect dropdown row's "tokens" modal lists exactly what to register.
• Generate fails with "No LLM provider configured": no Anthropic / OpenAI / Gemini key saved yet. Route to /viralman-login-gitmail, or have the user save one via the Gitmail row in the Connect dropdown.
• gitmail won't actually send: SMTP creds not configured. Same modal covers SMTP fields.
• Project disappears between page tabs: localStorage was cleared, or a different browser/profile is in use. Just retype.

Boundaries

• The skill never reads ~/.viralman/.env. Only viralman save-creds / / viralman post-* / viralman gitmail do.
• The skill never starts the server with --debug unless the user asked — the reloader can race with the OAuth state in session.
• The skill never bypasses the "I confirm" checkbox on the Gitmail page.