L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Productivité · SKILL.md

studio

studio

Claude World Studio CLI - full command reference for all CLI operations

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: studio description: Claude World Studio CLI - full command reference for all CLI operations user_invocable: false

Claude World Studio CLI

Full CLI for Claude World Studio. Maps 1:1 to the REST API + WebSocket.

Install

# Option A: npm global (recommended)
npm install -g @claude-world/studio
studio <command>

# Option B: From source
git clone https://github.com/claude-world/claude-world-studio.git
cd claude-world-studio && npm install
node bin/cli.js <command>
# or: npm link && studio <command>

# Option C: Electron Desktop App
npm run electron:dev            # development
npm run electron:build          # production (.app + .dmg)

MCP Server Setup

# One-command setup (uvx preferred, auto-cached)
npx @claude-world/studio setup-mcp

# Or legacy venv mode
npx @claude-world/studio setup-mcp --venv

# Update installed servers
npx @claude-world/studio setup-mcp --update

Quick Reference

# Server
studio serve                    # Start web UI
studio status                   # Check if running

# Sessions
studio session list
studio session create --title "Research" --workspace /path
studio session get <ID>
studio session rename <ID> --title "New Title"
studio session delete <ID>
studio session messages <ID> --limit 20

# Chat (WebSocket streaming)
studio chat --message "Find trends" --json
studio chat --session <ID> --message "Publish the best"
echo "What's trending?" | studio chat --json
studio interrupt <ID>

# Accounts
studio account list
studio account create --name "Main" --handle "@me" --platform threads --token TOKEN
studio account update <ID> --token NEW_TOKEN
studio account delete <ID>

# Settings
studio settings get
studio settings detect          # Scan system for MCP tools
studio settings apply           # Apply detected values
studio settings set --language en --default-workspace /path

# Publishing (supports all Threads post types)
studio publish --account <ID> --text "Hello!" --score 85
studio publish --account <ID> --text "Vote!" --poll "Option A|Option B"
studio publish --account <ID> --text "Look!" --image "https://url/img.png"
studio publish --account <ID> --text "Watch" --video "https://url/vid.mp4"
studio publish --account <ID> --text "Slides" --carousel URL1 URL2 URL3
studio publish --account <ID> --text "Mood" --gif-id "GIPHY_ID"
studio publish --account <ID> --text "Read" --link-attachment "https://url"
studio publish --account <ID> --text "TL;DR" --text-attachment article.txt
studio publish --account <ID> --text "Spoiler!" --image URL --spoiler-media
studio publish --account <ID> --text "Secret text" --spoiler-text "7:4"
studio publish --account <ID> --text "24hr only" --ghost
studio publish --account <ID> --text "My take" --quote-post-id "12345"
studio publish --account <ID> --text "Post" --reply-control mentioned_only
studio publish --account <ID> --text "Post" --topic-tag "AI" --alt-text "desc"
studio publish --account <ID> --text "Post" --link-comment "https://url"
studio history --limit 10

# Files
studio file list <SESSION_ID> --depth 2
studio file read <SESSION_ID> src/index.ts

Global Flags

FlagEnv VarDefaultDescription
--json—falseJSON output for all commands
--port NSTUDIO_PORT3001Server port
--host HSTUDIO_HOST127.0.0.1Server host

Chat Command

The chat command connects via WebSocket for real-time streaming.

Auto-create session: Omit --session to auto-create. Session ID is printed to stderr.

Input sources (priority order):

  1. --message "text" flag
  2. stdin pipe: echo "text" | studio chat

Output modes:

  • Default: assistant text → stdout, tool calls → stderr (with --verbose)
  • --json: NDJSON — one JSON object per event line

NDJSON event types:

{"type":"assistant_message","content":"Here are the trends..."}
{"type":"tool_use","toolName":"get_trending","toolId":"..."}
{"type":"tool_result","content":"..."}
{"type":"result","success":true,"cost_usd":0.03}

Ctrl+C sends an interrupt to the session and exits with code 130.

Production Pipeline Examples

Headless: Research → Publish

SESSION=$(studio session create --title "Auto Pipeline" --json | jq -r '.id')
studio chat --session $SESSION --message "Find top 3 trending AI topics in Taiwan" --json > trends.jsonl
RESPONSE=$(grep '"type":"assistant_message"' trends.jsonl | tail -1 | jq -r '.content')
studio publish --account acc123 --text "$RESPONSE" --score 80 --json

Freestyle: Full Autopilot

studio chat --message "Freestyle: find trending topics, research the best, \
  write a Threads post, score it, and publish to @mybrand" --json

Scheduled: Daily Content via Cron

# System cron (daily 9 AM)
0 9 * * * studio chat --message "Freestyle pipeline for @mybrand" --json >> ~/studio.log 2>&1

Or use Studio's built-in scheduler: Settings > Scheduled Tasks with per-account targeting and quality gates.

Multi-Account: Different Personas

studio publish --account brand-tech --text "Technical deep-dive..." --score 85
studio publish --account brand-casual --text "Hey did you hear..." --score 75

Settings Keys

CLI FlagAPI KeyDescription
--languagelanguageUI language (en, zh-TW, ja)
--themethemeUI theme (light, dark, system)
--trend-pulse-pythontrendPulseVenvPythontrend-pulse venv python path
--cf-browser-pythoncfBrowserVenvPythoncf-browser venv python path
--notebooklm-pathnotebooklmServerPathNotebookLM server.py path
--cf-browser-urlcfBrowserUrlCF Browser worker URL
--cf-browser-keycfBrowserApiKeyCF Browser API key
--cf-account-idcfAccountIdCloudflare account ID (cf-api mode)
--cf-api-tokencfApiTokenCloudflare API token (cf-api mode)
--default-workspacedefaultWorkspaceDefault workspace path

Instructions for Claude Code

When operating Studio via CLI:

  1. Always use --json for programmatic access
  2. Check status first: studio status --json before other commands
  3. Chat auto-creates sessions — no need to manually create unless reusing
  4. Parse NDJSON line by line for chat output
  5. Use stdin pipe for long messages or multi-line content
  6. Interrupt via CLI rather than killing the process: studio interrupt <ID>
Skills enregistrées