openclaw configure add more skills. fails.

[median trout]

In Terminal, I'm executing openclaw configure so I can add a couple more skills. However after I select "Configure skill now?" Yes, OpenClaw executes a config overwrite and I'm back to "What do you want to configure?". I'm not being presented the list of skills to choose from.

[hexed umbra]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[zenith fiber]

hippycm, first thing: what does openclaw --version show?

I checked the current docs/source, and this bit is confusing: openclaw configure -> skills -> Configure skills now? is mainly for checking already-visible skills, installing missing dependencies for those skills, and setting required env/API keys. It is not the ClawHub skill picker/install flow. If there are no installable missing requirements to prompt for, it can write the config and drop you back at “What do you want to configure?”

For adding more skills, use the skills CLI instead:

openclaw skills search "calendar"
openclaw skills install <skill-slug>
openclaw skills list --eligible
openclaw skills check

Then start a new OpenClaw session so the new workspace skill snapshot is picked up.

Docs:
https://docs.openclaw.ai/cli/skills

[zenith fiber]

https://docs.openclaw.ai/tools/skills
https://docs.openclaw.ai/tools/clawhub

If you expected openclaw configure itself to browse/install ClawHub skills, that may be a UX bug or missing path. Please include your openclaw --version, OS, and the exact skill you’re trying to add if you file it here: https://github.com/openclaw/openclaw/issues

[median trout]

OpenClaw 2026.5.12 (f066dd2)

[zenith fiber]

Thanks. On 2026.5.12, what you’re seeing matches the current flow: openclaw configure -> skills only configures/checks local/bundled skills and their missing dependencies. It doesn’t open a ClawHub browser/list to add new skills.

Use this path instead:

openclaw skills search "<what you want>"
openclaw skills install <skill-slug>
openclaw skills list --eligible
openclaw skills check

Example:

openclaw skills search "calendar"
openclaw skills install calendar

[zenith fiber]

After installing, start a new session so the skill snapshot reloads. If openclaw skills install ... fails, paste that exact output plus the skill slug and we can debug the actual install path.

Docs: https://docs.openclaw.ai/cli/skills

[median trout]

I actually want to enable an openclaw bundled skill.

[zenith fiber]

Got it. Bundled skills usually don’t need an “install” step: they’re loaded automatically unless they’re disabled, blocked by skills.allowBundled, or missing their declared requirements.

Can you run these and paste the relevant lines for the skill?

openclaw skills list --verbose
openclaw skills info <bundled-skill-name>
openclaw skills check
openclaw config get skills

Things to look for:

• If it says blocked by allowlist, add it to skills.allowBundled or remove that allowlist.
• If it says disabled, set skills.entries.<skillName>.enabled: true.
• If it shows missing bins/env/config, satisfy those requirements first.
• After changing config, start a new session so the skill snapshot reloads.

[zenith fiber]

Config shape is roughly:

{
  skills: {
    allowBundled: ["skill-name"],
    entries: {
      "skill-name": { enabled: true }
    }
  }
}

Docs: https://docs.openclaw.ai/tools/skills-config

If you tell me the exact bundled skill name, I can give the precise config key/path for that one.

[median trout]

summarize

[zenith fiber]

For the bundled summarize skill, the gate is simple: OpenClaw only loads it when the summarize CLI is on the Gateway host PATH.

Check it:

command -v summarize
openclaw skills info summarize
openclaw skills check

If command -v summarize is empty, install the CLI first. The bundled metadata advertises this installer:

brew install steipete/tap/summarize

Then make sure config isn’t blocking it:

[zenith fiber]

openclaw config get skills.allowBundled
openclaw config get skills.entries.summarize

If skills.allowBundled exists, summarize must be included in that array. If skills.entries.summarize.enabled is false, flip it:

openclaw config set skills.entries.summarize.enabled true

Then start a new OpenClaw session so the skill snapshot reloads.

Docs: https://docs.openclaw.ai/tools/skills-config

[median trout]

openclaw config set skills.entries.summarize.enabled true openclaw skills info summarize summarize △ Needs setup

Summarize or transcribe URLs, YouTube/videos, podcasts, articles, transcripts, PDFs, and local files.

Details:
Source: openclaw-bundled
Path: /opt/homebrew/lib/node_modules/openclaw/skills/summarize/SKILL.md
Homepage: https://summarize.sh
Visible to model: no
Available as command: no

Requirements:
Binaries: ✗ summarize

Install options:
→ Install summarize (brew)

[zenith fiber]

That output is the key bit:

Requirements:
  Binaries: ✗ summarize

So config is no longer the blocker. The bundled skill is enabled, but OpenClaw will keep it hidden until the summarize binary exists on the Gateway host PATH.

Run:

brew install steipete/tap/summarize
command -v summarize
openclaw skills info summarize

[zenith fiber]

If command -v summarize works in your terminal but openclaw skills info summarize still says Binaries: ✗ summarize, restart the gateway/app so the service picks up the updated PATH:

openclaw gateway restart

On macOS, service PATH can be the annoying part. If it still fails after restart, paste:

command -v summarize
openclaw status --all
openclaw skills info summarize

One more thing: once the skill loads, the summarize CLI itself still needs a provider key for the model it uses, commonly GEMINI_API_KEY for its default Gemini model, or another supported provider key.

[median trout]

How do I use the openclaw bundled summarize skill?

[zenith fiber]

Once it shows as eligible, you use the bundled skill by asking OpenClaw normally, or by invoking the generic skill command.

Setup first:

brew install steipete/tap/summarize
command -v summarize
openclaw gateway restart
openclaw skills info summarize

You want openclaw skills info summarize to stop saying Needs setup, and Visible to model should become yes.

Then use it in chat like:

[zenith fiber]

summarize this URL: https://example.com/article
what's this video about? https://youtu.be/...
transcribe this YouTube video: https://youtu.be/...
use summarize.sh on this PDF: /path/to/file.pdf

Or explicitly:

/skill summarize https://example.com/article
/skill summarize transcribe this YouTube video: https://youtu.be/...

The skill calls the external summarize CLI, so that CLI also needs a model provider key. Its default model is Gemini, so commonly:

[zenith fiber]

export GEMINI_API_KEY=...

For a daemon/gateway service, make sure that key is available to the gateway process too, not just your current terminal.

[median trout]

Thanks for your help. summarize is now enabled. How about the openai-whisper bundled skill?

[zenith fiber]

For the bundled openai-whisper skill, it’s the same pattern as summarize, but the required binary is whisper.

Local/offline-ish Whisper CLI skill:

brew install openai-whisper
command -v whisper
openclaw gateway restart
openclaw skills info openai-whisper

You want openclaw skills info openai-whisper to stop saying Needs setup, and Visible to model should become yes.

Then use it in chat:

[zenith fiber]

transcribe this audio file with whisper: /path/to/audio.mp3
translate this audio to English with whisper: /path/to/audio.m4a
/skill openai-whisper transcribe /path/to/audio.mp3

A couple notes:

• First run downloads Whisper models into ~/.cache/whisper.
• The skill metadata says whisper is the required binary; in practice, audio decoding may also need ffmpeg, so install that if Whisper errors on your file format: brew install ffmpeg.
• This is different from openai-whisper-api, which uses OpenAI’s hosted transcription API and requires OPENAI_API_KEY instead of the local whisper binary.

Docs: https://docs.openclaw.ai/tools/skills-config

[median trout]

I actually meant openai-whispeer-api

[zenith fiber]

Yep — that one is openai-whisper-api (one p in whisper). It uses OpenAI’s hosted audio transcription API, not the local Whisper CLI.

Requirements from the bundled skill:

Binaries: curl
Env: OPENAI_API_KEY

Check status:

command -v curl
openclaw skills info openai-whisper-api

Set the API key for the gateway. The cleanest config path is the skill entry API key:

[zenith fiber]

openclaw config set skills.entries.openai-whisper-api.apiKey --ref-provider default --ref-source env --ref-id OPENAI_API_KEY

Then make sure OPENAI_API_KEY is available to the gateway process, restart, and re-check:

openclaw gateway restart
openclaw skills info openai-whisper-api

Once eligible, use it like:

/skill openai-whisper-api transcribe /path/to/audio.m4a
/skill openai-whisper-api transcribe /path/to/audio.m4a --language en

[zenith fiber]

/skill openai-whisper-api transcribe /path/to/audio.m4a --prompt "Speaker names: Alice, Bob"

Or ask naturally: “transcribe this audio via OpenAI Whisper API: /path/to/audio.m4a”.

If you use an OpenAI-compatible proxy/local gateway, set OPENAI_BASE_URL for the gateway process too.

[median trout]

Back to the openai-whisper skill. openclaw skills info openai-whisper 🎤 openai-whisper ⏸ Disabled

Local speech-to-text with the Whisper CLI (no API key).

Details:
Source: openclaw-bundled
Path: /opt/homebrew/lib/node_modules/openclaw/skills/openai-whisper/SKILL.md
Homepage: https://openai.com/research/whisper
Visible to model: no
Available as command: no

Requirements:
Binaries: ✓ whisper

Install options:
→ Install OpenAI Whisper (brew)

[zenith fiber]

That one is different now: the binary gate is satisfied.

Binaries: ✓ whisper
Status: Disabled

So enable the skill entry:

openclaw config set skills.entries.openai-whisper.enabled true
openclaw config get skills.entries.openai-whisper
openclaw gateway restart
openclaw skills info openai-whisper

After that, it should no longer show ⏸ Disabled, and Visible to model / Available as command should become available for new sessions.

[zenith fiber]

If it still shows disabled after that, paste:

openclaw config get skills
openclaw skills info openai-whisper

The Install OpenAI Whisper (brew) line is just the available installer metadata; since Binaries: ✓ whisper, you don’t need to run the installer again.

[median trout]

How do I use the openclaw openai-whisper skill with my OpenClaw Discord bot?

[zenith fiber]

With the Discord bot, use it on an uploaded audio attachment or Discord voice-message attachment, not a live voice channel.

Checklist first:

openclaw skills info openai-whisper
command -v whisper
command -v ffmpeg

You want the skill eligible/visible. If ffmpeg is missing, install it too:

brew install ffmpeg
openclaw gateway restart

[zenith fiber]

Then in Discord, attach the audio file and send something like:

transcribe this attached audio with openai-whisper

or explicitly:

/skill openai-whisper transcribe the attached audio

For translation:

translate this attached audio to English with openai-whisper

[zenith fiber]

OpenClaw should pass the attachment into the agent context as a local media path, and the skill tells the agent to run the local whisper CLI against that file. If it doesn’t pick it up, try a fresh Discord thread/session after enabling the skill, because skills are snapshotted per session.

Important distinction: this is for audio files/voice-message attachments. For Discord realtime voice channels, use OpenClaw’s Discord voice features instead; the openai-whisper skill is just local file transcription.

[median trout]

How do I use OpenClaw's Discord voice feature?

[zenith fiber]

OpenClaw’s Discord voice feature is for realtime voice-channel conversations. It’s separate from audio attachments / the Whisper skill.

Minimum setup:

openclaw config set channels.discord.voice.enabled true
openclaw config set channels.discord.commands.native true
openclaw gateway restart

In Discord, make sure the bot has permissions in the target voice channel:

• View Channel
• Connect
• Speak

Then use the Discord native slash command:

[zenith fiber]

/vc status
/vc join
/vc leave

Important: /vc is a Discord native command, not a normal text command, so it needs native command registration enabled and the gateway restarted after config changes.

Once joined, talk in the voice channel. OpenClaw should transcribe speaker turns, run them through the default Discord/account agent, and speak replies back using the configured voice TTS. Voice turns follow the same allowlist/group-policy rules as normal Discord commands, so non-owner speakers should not get owner-only tools.

Useful checks if it doesn’t join or can’t hear/speak:

openclaw channels status --probe
openclaw doctor
openclaw logs --follow

[zenith fiber]

command -v ffmpeg
command -v ffprobe

Docs: https://docs.openclaw.ai/channels/discord#voice