Fix duplicate reviewer assignment in assign-reviews workflow

[neubig]

Summary

Fixes #826

The assign-reviews.yml workflow was assigning reviewers to PRs that already had reviewers assigned, causing duplicate assignments. This happened because the prompt was not explicit enough about how to check for existing reviewers using the GitHub API.

Changes

This PR updates the "Need Reviewers" section of the workflow prompt to:

1. Add explicit GitHub API instructions: Clearly specify that the agent should use GET /repos/{owner}/{repo}/pulls/{pull_number} to fetch PR details
2. Check both reviewer arrays: Explicitly check both requested_reviewers and requested_teams arrays
3. Include concrete examples: Provide actual API response examples showing:
   • A PR that DOES NOT need reviewers (has entries in the arrays - should be skipped)
   • A PR that DOES need reviewers (empty arrays - should be processed)
4. Emphasize the skip condition: Make it clear that if either array has any entries, the PR should be skipped

Example

The updated prompt now includes concrete examples like:

Example API response showing a PR that DOES NOT need reviewers (skip this):
{
  "requested_reviewers": [{"login": "someuser"}],
  "requested_teams": []
}

Example API response showing a PR that DOES need reviewers (process this):
{
  "requested_reviewers": [],
  "requested_teams": []
}

Testing

This is a configuration change to a GitHub workflow file. The change will be validated when the workflow runs next (scheduled daily at 12 PM UTC or can be triggered manually).

Impact

This change should prevent the workflow from:

• Assigning duplicate reviewers to PRs that already have reviewers
• Creating redundant assignment comments on PRs

The agent will now properly identify PRs that truly have no reviewers assigned and skip those that already have reviewers.

@neubig can click here to continue refining the PR

---

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Base Image	Docs / Tags
golang	golang:1.21-bookworm	Link
java	eclipse-temurin:17-jdk	Link
python	nikolaik/python-nodejs:python3.12-nodejs22	Link

Pull (multi-arch manifest)

docker pull ghcr.io/openhands/agent-server:f41a0db-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-f41a0db-python \
  ghcr.io/openhands/agent-server:f41a0db-python

All tags pushed for this build

ghcr.io/openhands/agent-server:f41a0db-golang
ghcr.io/openhands/agent-server:v1.0.0a3_golang_tag_1.21-bookworm_binary
ghcr.io/openhands/agent-server:f41a0db-java
ghcr.io/openhands/agent-server:v1.0.0a3_eclipse-temurin_tag_17-jdk_binary
ghcr.io/openhands/agent-server:f41a0db-python
ghcr.io/openhands/agent-server:v1.0.0a3_nikolaik_s_python-nodejs_tag_python3.12-nodejs22_binary

The f41a0db tag is a multi-arch manifest (amd64/arm64); your client pulls the right arch automatically.

[blacksmith-sh]

[Automatic Post]: This PR seems to be currently waiting for review. @simonrosenberg, could you please take a look when you have a chance?

[enyst]

This one seems a bit tricky, it's also possible to review a part of a PR and not the rest? 🤔

I'm not sure what is best:

• PRs with two reviewers
• PRs with a comment, and no more response/follow-up review

Probably two reviewers? 🤔

We can try it like this, see how it works, but idk, we might want to revisit.

[enyst]

I have a little doubt, but maybe we can try this prompt and see how we like the effect! Not sure how Simon thinks about it?

[enyst]

Looks like this PR is needed:

• Rename metadata field to litellm_extra_body and add custom config support #837 (comment)

In that case, the PR

• had a reviewer (assigned)
• it was approved, too
• and yet, another reviewer was assigned.