How to use the Webhook Event Transforms agent skill

What the skill makes the agent do

Agent skills are instruction packages that Claude can load to perform a specific kind of task. This one covers the path from an external event to a brain page. You define a transform function that maps the fields of an incoming event to a page, register a webhook URL with the external service, and then the agent processes each event as it arrives.

For each event the agent parses the payload, runs your transform, writes the page with gbrain put, extracts entities, adds timeline entries to mentioned entities, and runs gbrain sync. The output is markdown content plus metadata such as slug, type and citations.

• SMS: a timeline entry on the sender's brain page, plus a task update if an action item is detected.
• Finished meeting: delegates to the meeting-ingestion skill.
• Social mention: a brain page in media/, with entity extraction and backlinks.

Install the webhook-transforms skill

Download the ZIP for this skill from this page. The skill package contains a single file, SKILL.md.

Option one is a personal install: unzip the download into ~/.claude/skills/ so the webhook-transforms folder sits there. Option two is a project install: unzip it into the .claude/skills/ folder inside your project.

Option three is to upload the ZIP in the Claude apps under Settings, then Capabilities, then Skills.

Check the requirements before you start

The skill only works with a webhook-fed pipeline and a running gbrain installation. Before setup, confirm that gbrain is installed and configured with the put_page, add_timeline_entry and search tools.

You also need a brain host that issues a token with write scope for the target source, and an external service that can send webhooks to an endpoint you control. The skill itself runs no scripts, but it does use the network.

• gbrain installed and configured with put_page, add_timeline_entry and search.
• A brain host token with write scope for the target source.
• An external service able to send webhooks to an endpoint you control.

Trigger the skill with natural requests

The skill loads when you ask for one of its trigger phrases. Use wording close to these: set up webhook, process webhook event, or transform this event. Say what the event is and where it should go, so the agent can choose the right transform.

Describe the event type in plain words. For example, you can ask the agent to set up a webhook for incoming text messages, or to transform this event from a meeting tool into a brain page.

• Set up webhook for incoming SMS messages.
• Process webhook event from my meeting recorder.
• Transform this event: a social mention from a platform.

Understand the safety rules and failure handling

External input is written into a persistent store, so the skill sets safety rules. Transforms must strip HTML tags and escape script content, so no raw HTML or script reaches a brain page. Every transformed event gets citations and entity extraction.

If a transform throws an error, the skill writes the raw payload to a dead-letter file under _dead-letter/ with a timestamp, reports the error type, and retries once. A malformed payload is rejected with its parse error and is not written as a partial page.

If a write returns write_pending, the event is accepted and the agent records the request id and polls it, rather than replaying the event under a new id. If the token lacks scope for a source, the agent tells you which grant the brain host operator must add.

• Do not expose an endpoint until you have reviewed sanitization and token scopes.
• Inaccurate or hostile payloads can pollute the store, because writes persist.