Skip to content

Getting started

This guide gets crab’d running on a repository in a few minutes. You’ll add a workflow, give it a model key, and mention crab’d on an issue.

  • A repository on GitHub or Forgejo.
  • An API key for at least one model provider (for example ANTHROPIC_API_KEY or OPENAI_API_KEY), stored as a CI secret.

Create .github/workflows/crabd.yml (or .forgejo/workflows/crabd.yml on Forgejo):

.github/workflows/crabd.yml
name: crab'd
on:
issue_comment:
types: [created]
pull_request:
types: [opened, reopened, ready_for_review]
issues:
types: [assigned, labeled]
jobs:
crabd:
runs-on: ubuntu-latest
permissions:
contents: write
pull-requests: write
issues: write
id-token: write
steps:
- uses: actions/checkout@v4
- uses: louisescher/crabd@v1
with:
model: anthropic/claude-sonnet-4-6
providers: anthropic
env:
ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

Once the workflow is committed, try any of these:

  • Mention — comment /crabd how does the auth middleware work? on an issue or PR.
  • Review — open a pull request; crab’d reviews the diff automatically.
  • Implement — assign an issue to crab’d (or label it) to have it open a PR.

crab’d posts a “working…” comment, then updates it in place with the result.

Anything you write after the mention is passed to the agent, so you can steer any mode:

/crabd review — focus on the database migration and error handling
  • Configuration — every dial, layered config, and governance.
  • Providers — use OpenAI, OpenRouter, or a local model.
  • Modes — how mention, review, and implement behave.