GitHub
Ask Escher to do something in GitHub. Escher classifies your request, generates the Python script that does it, and runs it from your machine.
How it works
GitHub is one of the targets recognised by Escher's integrations agent. When you type a GitHub-related request in chat, Escher:
- Classifies the integration (Linear / Jira / Slack / GitHub / unknown) using a small classifier model
- Routes the query to the code-gen agent with
action: "github" - Streams back a generated Python script for the operation
- The script runs from your machine using your GitHub credentials
This is the same pattern Escher uses for Linear and Jira. Escher does not maintain a long-lived connection to your GitHub account or webhook subscriptions — each request produces a fresh script that uses GitHub's REST/GraphQL APIs.
What you can ask
Phrasing that classifies cleanly to GitHub:
"List all open PRs assigned to me in
escher-dbai."
"Find issues in
escher-documentationtaggedbugand unassigned."
"Comment on PR #847 with a status update from yesterday's deploy."
The richer the GitHub API supports, the richer Escher's generated script can be — Escher generates against the standard GitHub APIs.
Authentication
The generated script reads your GitHub credentials from your environment (typically a personal access token in GITHUB_TOKEN or your gh CLI session). Set whichever your team uses before running.
What's not supported
Repo evidence today only covers natural-language → script generation for GitHub. Anything beyond that — saved notification rules, deploy-to-cloud-event correlation, pre-merge risk scoring, PR-linked Canvases — is not currently shipped.