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:
reusededited_heavilyignoredconfusing
And one AI review fit answer:
in_additionbefore_reviewinsteadnot_neededunsure
-
Copy examples/github-actions/relay-external-trial.yml into the target repository as:
.github/workflows/relay-external-trial.yml -
Open or update a pull request in that repository.
-
Let the
Relay External Trialworkflow run. -
Download the
relay-validation-bundleartifact. -
Compare
.relay/validation-brief.mdor.relay/pr-comment.mdwith the normal PR summary. -
File feedback through the Round 1 Relay feedback form.
The workflow:
- checks out the target repository
- downloads
relay_runtime.pyfrom thev0.1.13release 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-bundleartifact
The bundle includes:
.relay/review-readiness.md.relay/handoff.md.relay/pr-comment.md.relay/reviewer-pack.md.relay/validation-brief.mdrelay-review-readiness.jsonrelay-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
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.
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.
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
If the target repository does not want to keep Relay:
- Delete
.github/workflows/relay-external-trial.yml. - Do not commit
.relay/artifacts. - Keep only the linked feedback issue as the evidence record.
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. |