1. First attempt — failed (permissions)

curl -fsSL https://openclaw.ai/install.sh | bash

EACCES: permission denied — old install owned directory as root.

2. Fix permissions

sudo chown -R $(whoami) /usr/local/lib/node_modules/

3. Success

curl -fsSL https://openclaw.ai/install.sh | bash

v2026.2.25 installed. Gateway token auth configured, LaunchAgent installed.

OpenAI Codex (default)

openclaw configure → Model → OpenAI → OpenAI Codex (ChatGPT OAuth)

Anthropic (added later)

Via env.ANTHROPIC_API_KEY in openclaw.json. Auto-detected.

Available: anthropic/claude-opus-4-6, anthropic/claude-sonnet-4-6

openclaw configure → Channels → Slack (Socket Mode)

• Bot token: xoxb-[redacted]
• App token: xapp-[redacted]
• Channel allowlist: [lobster-test],[video-production]
• DM policy: Pairing — approved sender [redacted]

Default Slack bot. Handles all channels except #video-production.

SOUL.md + IDENTITY.md (lobster)

AI video production for Punchlist Zero.

SOUL.md + IDENTITY.md + config/{technical_notes, style_guide, demo_context}.md

Gotcha: Agents go in agents.list[] array, NOT top-level keys. First attempt was rejected.

{
  "bindings": [
    {
      "agentId": "video-producer",
      "match": { "channel": "slack", "peer": { "kind": "channel", "id": "C0AGT89CSKD" } }
    },
    {
      "agentId": "lobster",
      "match": { "channel": "slack" }
    }
  ]
}

Order matters: specific peer match first, catch-all second.

Verified 2026-02-26 05:40 UTC

7-step human-in-the-loop. Every step needs Dustin's approval.

Green = built skill

Agent itself (Opus 4.6). Topic → 3 angles → script → shot-by-shot storyboard in Slack thread.

pexels-stock + pixabay-stock skills. Search, preview, download. 1080p+ filter baked in.

sag skill. ElevenLabs TTS, Roger voice. Preview via MEDIA: prefix.

pillow-video-cards + ffmpeg-assembler. Section cards via Pillow, then assemble.py --manifest → final MP4.

pillow-video-cards — thumbnail.py. 1280x720 PNG, bg-color or bg-image.

youtube-upload skill. Resumable upload via YouTube Data API v3.

scripts/ search_video.py, search_photo.py, download.py, preview.py

Requires PEXELS_API_KEY + requests. API was 503 at test (transient).

scripts/ section_card.py, intro_slide.py, outro_slide.py, thumbnail.py, text_overlay.py

All 5 tested. Brand: Gold #F5B800, Arial Bold, Impact.

scripts/ assemble.py, normalize.py, concat_clips.py, mix_audio.py, overlay_image.py

normalize.py tested. Full assembly pending real media.

scripts/ search_video.py, search_photo.py, download.py

500 results for "solar energy". stdlib urllib only.

scripts/ upload.py, set_thumbnail.py, check_quota.py
references/ youtube_api.md

Resumable upload (10MB chunks). OAuth active — Testing_Gluthoric (UCv0sMGwhWthhXT715pepuxg).

ClawHub reference skill. API guides for Runway Gen-4.5, Luma, Kling, Pika, Replicate, Stable Video Diffusion. Agent reads docs and writes API calls.

scripts/ generate_srt.py

Whisper transcription + SRT generation. Supports --burn for hardcoded subtitles, --translate for language translation.

• No ffmpeg drawtext — no freetype. ALL text via Pillow.
• Full binary paths — /usr/local/bin/{ffmpeg,python3.12,sag}
• File uploads — MEDIA:/path/to/file in agent replies
• Normalize clips — 1920x1080@30fps before assembly
• Roger voice only — CwhRBWXzGAHq8TQ4Fs17
• Runway ML broken — expired. Pexels + Pixabay only.

~/.openclaw/workspace-video/
├── SOUL.md, IDENTITY.md
├── config/{technical_notes,style_guide,demo_context}.md
├── pexels-stock/          # Stock footage
├── pillow-video-cards/    # Text graphics
├── ffmpeg-assembler/      # Video assembly
├── pixabay-stock/         # Backup stock
├── youtube-upload/        # YouTube API (tested)
├── assets/{stock_video,stock_images,music,voiceover}/
└── output/{videos,thumbnails,metadata}/

The video pipeline needs 6 API keys from different providers. Two of them (Anthropic & OpenAI) are configured via openclaw configure OAuth login. The other four are pasted into ~/.openclaw/openclaw.json as environment variables.

Time estimate: ~30 minutes total. Pexels and Pixabay are instant. ElevenLabs takes 2 minutes. Anthropic and OpenAI each take ~5 minutes. YouTube/GCP is the longest at ~15 minutes due to the OAuth flow.

What it's for: Powers the video-producer agent (Claude Opus 4.6). This is the brain that writes scripts, plans storyboards, and orchestrates the entire video pipeline.

1. Go to console.anthropic.com
2. Click "Continue with Google" or enter your email
3. Verify your email if signing up new

1. In the left sidebar, click Billing
2. Add a credit card (you only pay for what you use)
3. Set a monthly spend limit if you want (recommended: start with $20)

Typical cost per video: ~$0.50-2.00 depending on script length and revisions.

1. Left sidebar: API keys (under MANAGE)
2. Click "+ Create Key" (top right)
3. Name it something like openclaw-travis
4. Leave workspace as Default
5. Click Add
6. Copy the key immediately — it starts with sk-ant-api03- and you won't see it again

The Create API Key dialog. Pick a name, leave workspace as Default, click Add.

Save it: Paste the key somewhere safe (password manager, secure note). You'll need it for the openclaw configure step later.

What it's for: Powers the lobster agent (GPT-5.3-Codex). This is the general-purpose Slack bot that handles everything except video production.

1. Go to platform.openai.com
2. Sign up or log in with your existing ChatGPT account

1. Top-left dropdown: click your org name → Billing
2. Add a payment method
3. Set a monthly budget limit (recommended: start with $20)

1. Left sidebar: API keys (under Manage)
2. Click "+ Create new secret key" (top right)
3. Owned by: You
4. Name: openclaw-travis
5. Project: Default project
6. Permissions: All
7. Click "Create secret key"
8. Copy the key immediately — it starts with sk- and you won't see it again

The Create Secret Key dialog. Set name, leave permissions as All, click Create secret key.

Alternative: OpenClaw also supports OpenAI Codex OAuth login (the ChatGPT-based login). During openclaw configure you can choose "OpenAI Codex" and log in with your ChatGPT account instead of pasting an API key. This is what we use on the Momenta Mac.

What it's for: Text-to-speech voice synthesis. The video pipeline uses ElevenLabs to generate narration audio using the "Roger" voice (CwhRBWXzGAHq8TQ4Fs17).

1. Go to elevenlabs.io
2. Click "Sign Up" (top right) — Google sign-in works
3. The free tier gives you 10,000 characters/month (enough for ~2-3 short videos)

1. Click your profile icon (bottom-left) → Developers
2. Or go directly to: elevenlabs.io/app/developers/api-keys
3. Click the "API Keys" tab
4. Click "+ Create Key"
5. Your key will appear in the list — click the copy icon next to it
6. Make sure the key's Enabled toggle is ON (dark toggle = enabled)

The ElevenLabs Developers → API Keys page. Click + Create Key, then toggle it on.

Free tier limits: 10,000 characters/month. For heavier use, the Starter plan ($5/mo) gives 30,000 characters. The video pipeline uses ~1,500-3,000 characters per video narration.

What it's for: Free stock photos and videos. The pexels-stock skill searches and downloads HD/4K footage for video B-roll.

1. Go to pexels.com
2. Click "Join" (top right) — Google sign-in works

1. Go to pexels.com/api
2. Click "Your API Key"
3. If this is your first time, you'll fill out a short form:
   • What are you building? → "AI video production tool"
   • URL → can be anything (your website, GitHub, etc.)
4. After submitting, your key appears on the page at pexels.com/api/key
5. Copy the long alphanumeric string from the text box

Completely free. 200 requests/hour, 20,000 requests/month. No credit card needed. If you need more, email api@pexels.com.

What it's for: Backup stock photos and videos. The pixabay-stock skill is used when Pexels is down or doesn't have the right footage.

1. Go to pixabay.com
2. Click "Join" (top right)
3. Sign up with email or Google

1. Go to pixabay.com/api/docs
2. If you're logged in, your API key is shown directly on the page in a green highlighted box next to "Your API key:"
3. It's in the "Parameters" section under "Search Images" — scroll down to find it
4. Copy the key (it's a long alphanumeric string like 12345678-abcdef...)

Completely free. 100 requests per 60 seconds. No credit card needed. Results must be cached for 24 hours per their terms.

What it's for: Uploading finished videos to YouTube via the Data API v3. This is the most complex setup because it requires creating a Google Cloud project and going through an OAuth flow.

This is optional. You only need this if you want the bot to upload videos directly to YouTube. You can skip this and upload manually if you prefer.

1. Go to console.cloud.google.com
2. Sign in with the Google account you want to use
3. Click the project dropdown (top bar, next to "Google Cloud")
4. Click "New Project"
5. Name: openclaw-yt-travis
6. Click Create
7. Wait ~30 seconds, then select the new project from the dropdown

1. In the left sidebar: APIs & Services → Library
2. Search for "YouTube Data API v3"
3. Click it, then click Enable

1. Left sidebar: APIs & Services → OAuth consent screen
2. User type: External → Create
3. Fill in the required fields:
   • App name: OpenClaw Video Upload
   • User support email: your email
   • Developer contact: your email
4. Click Save and Continue through the remaining steps (Scopes, Test Users, Summary)
5. On the Test Users step, add your own Gmail/Google email as a test user

1. Left sidebar: APIs & Services → Credentials
2. Click "+ Create Credentials" → OAuth client ID
3. Application type: Desktop app
4. Name: openclaw-desktop
5. Click Create
6. A dialog appears with your Client ID and Client Secret — copy both

Run the OAuth flow on your machine to get a refresh token:

pip3 install google-auth google-auth-oauthlib google-api-python-client

Create a file called get_token.py:

from google_auth_oauthlib.flow import InstalledAppFlow

flow = InstalledAppFlow.from_client_config(
    {
        "installed": {
            "client_id": "YOUR_CLIENT_ID",
            "client_secret": "YOUR_CLIENT_SECRET",
            "auth_uri": "https://accounts.google.com/o/oauth2/auth",
            "token_uri": "https://oauth2.googleapis.com/token"
        }
    },
    scopes=["https://www.googleapis.com/auth/youtube.upload",
            "https://www.googleapis.com/auth/youtube"]
)
creds = flow.run_local_server(port=8080)
print(f"\nRefresh Token: {creds.refresh_token}")
print(f"Access Token:  {creds.token}")

Run it:

python3 get_token.py

1. A browser window opens — sign in with your YouTube account
2. Click "Continue" through the consent screen (it'll warn about unverified app — click "Advanced" → "Go to OpenClaw Video Upload")
3. Grant access
4. Copy the Refresh Token from the terminal output

Important: The app stays in "Testing" mode, which means tokens expire every 7 days. You'll need to re-run get_token.py weekly, or submit your app for Google verification (takes days/weeks). For testing this is fine.

Now that you have all your keys, let's plug them into OpenClaw.

Run the interactive configurator:

openclaw configure

Navigate to: Model → Anthropic

• Choose "API Key" when prompted
• Paste your Anthropic API key (sk-ant-api03-...)

Then navigate to: Model → OpenAI

• Choose "OpenAI Codex (ChatGPT OAuth)" for the easiest login
• A browser window opens — log in with your ChatGPT account
• Or choose "API Key" and paste your sk-... key

Open the OpenClaw config file:

nano ~/.openclaw/openclaw.json

Add or update the env section with your keys:

{
  "env": {
    "ANTHROPIC_API_KEY": "sk-ant-api03-YOUR-KEY-HERE",
    "ELEVENLABS_API_KEY": "YOUR-ELEVENLABS-KEY",
    "PEXELS_API_KEY": "YOUR-PEXELS-KEY",
    "PIXABAY_API_KEY": "YOUR-PIXABAY-KEY",
    "YOUTUBE_CLIENT_ID": "YOUR-GCP-CLIENT-ID.apps.googleusercontent.com",
    "YOUTUBE_CLIENT_SECRET": "YOUR-GCP-CLIENT-SECRET",
    "YOUTUBE_REFRESH_TOKEN": "YOUR-REFRESH-TOKEN"
  }
}

Only add the keys you have. The file already has other settings — just add/update the env block.

Security: Never commit openclaw.json to git. Never share your API keys in Slack, email, or screenshots. If a key leaks, rotate it immediately on the provider's dashboard.

After configuring all keys, run the OpenClaw doctor to verify:

openclaw doctor

You should see green checks for all configured providers. Then test each one:

# In Slack, message the bot:
@lobster what model are you using?

# Or test directly:
curl -s https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-4-6","max_tokens":10,"messages":[{"role":"user","content":"hi"}]}' \
  | head -c 200

curl -s "https://api.elevenlabs.io/v1/user" \
  -H "xi-api-key: $ELEVENLABS_API_KEY" | python3 -m json.tool

Should return your account info with character limits.

curl -s "https://api.pexels.com/v1/search?query=solar+energy&per_page=1" \
  -H "Authorization: $PEXELS_API_KEY" | python3 -m json.tool

Should return a JSON response with photo results.

curl -s "https://pixabay.com/api/?key=$PIXABAY_API_KEY&q=solar+energy&per_page=3" \
  | python3 -m json.tool

Should return a JSON response with totalHits and image results.

# Uses the youtube-upload skill's check_quota.py:
python3 ~/.openclaw/workspace-video/youtube-upload/scripts/check_quota.py

Should show your channel name and remaining quota.

All set! Once all tests pass, the video pipeline is ready. Try sending a message to #video-production in Slack to kick off your first video.