Skip to content

Latest commit

 

History

History
130 lines (90 loc) · 4.01 KB

File metadata and controls

130 lines (90 loc) · 4.01 KB

External Repo Trial

Use this when a maintainer wants to try Relay on a real pull request without installing the Codex App plugin or committing .relay/ files.

The goal is not adoption yet. The goal is one outside reviewer outcome:

  • reused
  • edited_heavily
  • ignored
  • confusing

And one AI review fit answer:

  • in_addition
  • before_review
  • instead
  • not_needed
  • unsure

Fast Path

  1. Copy examples/github-actions/relay-external-trial.yml into the target repository as:

    .github/workflows/relay-external-trial.yml
    
  2. Open or update a pull request in that repository.

  3. Let the Relay External Trial workflow run.

  4. Download the relay-validation-bundle artifact.

  5. Compare .relay/validation-brief.md or .relay/pr-comment.md with the normal PR summary.

  6. File feedback through the Round 1 Relay feedback form.

What The Workflow Does

The workflow:

  • checks out the target repository
  • downloads relay_runtime.py from the v0.1.13 release tag and verifies its SHA256 checksum
  • runs review-readiness --base-ref origin/<base-branch>
  • runs validation-brief --base-ref origin/<base-branch>
  • writes a review-readiness summary to the GitHub Actions job summary
  • uploads a relay-validation-bundle artifact

The bundle includes:

  • .relay/review-readiness.md
  • .relay/handoff.md
  • .relay/pr-comment.md
  • .relay/reviewer-pack.md
  • .relay/validation-brief.md
  • relay-review-readiness.json
  • relay-validation-brief.json

It does not:

  • post comments
  • request reviews
  • block merges
  • send repository contents to a Relay server
  • require a Relay API key
  • require installing the Codex App plugin

Why It Downloads From A Tag

The example pins the runtime to:

https://raw.githubusercontent.com/Ryan-Guo123/relay-codex/v0.1.13/plugins/relay-codex/scripts/relay_runtime.py

Pinning to a tag is more reviewable than running from main. Maintainers should still inspect the workflow and runtime before enabling it on sensitive repositories.

The example also checks the downloaded runtime against:

242e631d6c98a205cfc88fe40620a7c6b96c89117da58336de9d4f094113fd9b

If Relay publishes a new runtime version, update both the tag and checksum in the copied workflow.

What To Ask The Reviewer

Ask one maintainer or frequent reviewer:

Could you compare the Relay validation bundle with the normal PR summary?

Please answer two things:

1. Outcome: reused, edited_heavily, ignored, or confusing.
2. AI review fit: in_addition, before_review, instead, not_needed, or unsure compared with Copilot code review, CodeRabbit, Graphite, Qodo, PR-Agent, or another AI review tool you use.

Negative feedback is useful. If an existing review tool already solves this job, mark Relay not_needed.

What Counts As Evidence

Count the trial only if:

  • the target repository is not Ryan-Guo123/relay-codex
  • the workflow ran on a real pull request or review task
  • the reviewer saw the generated artifact
  • the reviewer compared it with a normal PR summary
  • the result is recorded in a public feedback issue or linked public comment

Do not count:

  • copying the workflow without running it
  • this repository running its own Review Readiness workflow
  • a private compliment without one of the required outcomes
  • a star without a linked trial

Cleanup

If the target repository does not want to keep Relay:

  1. Delete .github/workflows/relay-external-trial.yml.
  2. Do not commit .relay/ artifacts.
  3. Keep only the linked feedback issue as the evidence record.

Product Decision

If the first external repo trial returns:

Result Decision
reused or edited_heavily Test one more external PR before building new features.
ignored or confusing Simplify the generated artifact before more outreach.
not_needed versus AI review tools Shrink Relay to a template/schema or stop the product direction.