An OpenFactory agent triaging issues and reviewing pull requests in connected GitHub repositories

Dev Integration

GitHub connection with repository-level guardrails

GitHub is available now; provider scope, exact account assignment, write approval, and upstream audit verification remain essential.

GitHub is configured in the production connection catalog. Connect a least-privileged test identity first, limit repository access upstream, bind the exact connection to one role, and require human approval for merges, releases, access changes, secrets, and destructive writes.

A concrete use case

Consider a repository-triage workflow. A new issue comes in with a vague title and no version info. Your agent reads it, labels it as a bug, and posts a friendly comment asking for the missing details and a repro. It links a related open issue it found. When a pull request lands, the same agent leaves a one-comment summary of what changed and flags that a migration file was touched without a test. The expected output is review material and, where approved, narrowly scoped comments or labels. Verify every write in GitHub's history.

Example prompt

Review and adapt this in console.openfactory.tech after connecting a test identity and limiting its repository access.

You are my repository triage agent with access to GitHub.

When a new issue is opened in my repo:
- Read it and label it (bug, feature, question, or needs-info).
- If key details are missing, post a friendly comment asking for them.
- Link any duplicate or related issues you find.

When a pull request is opened:
- Summarize what it changes in one comment.
- Flag obvious risks: missing tests, broad diffs, or touched migration files.

How to connect GitHub

  1. Open the OpenFactory console and choose GitHub.
  2. Authorize a dedicated identity and limit repositories and scopes in GitHub.
  3. Assign the exact connection to one role; test denied repositories, inspect audit history, and verify disconnection plus upstream revocation before enabling writes.

GitHub pairs naturally with issue tracking and chat - see all integrations for the current catalog and clearly labeled planned patterns.

Frequently asked questions

What can an AI agent do in GitHub?

The connected GitHub toolkit can expose issue, pull-request, and repository actions subject to its current scopes and tool catalog. Confirm each requested action in GitHub and keep merges, releases, secrets, access changes, and destructive operations outside autonomous approval.

Does the agent act under my identity?

Successful writes use the authorized GitHub connection. Confirm the displayed actor and event in GitHub's own history because attribution can vary with OAuth apps, installations, and token type.

Can I limit it to specific repositories?

Limit access using GitHub's own authorization or installation controls, then assign the exact connection to an OpenFactory role. Test a denied repository explicitly; a prompt or role name alone does not prove isolation.

Where does the connection live?

The managed GitHub path uses Composio Cloud for OAuth and token storage; OpenFactory also has a separate native GitHub OAuth path for repository delivery. Code or metadata used by an agent may traverse the selected connection, OpenFactory, and model path. Self-hosting the control plane alone does not make every path local.

Build the image instead of hand-assembling it

Use OpenFactory to turn the same requirements into a bootable, testable Linux system.

Open console