Skip to content
ReviewPhin Docs

Merge requests

GitLab merge requests are the primary ReviewPhin workflow. You drive everything from merge request comments and the merge request description.

Trigger a review

Section titled “Trigger a review”

Comment on the merge request, mentioning the bot:

@reviewphin review this

Use the bot username discovered from the GitLab access token. The first run is a full review of all changed files; later runs are incremental.

Force a full re-scan

Section titled “Force a full re-scan”

To ignore prior incremental context and rescan more broadly:

@reviewphin full review

Other accepted phrasings: full rescan, fresh full review, full review from scratch, rescan everything.

Ask about an image

Section titled “Ask about an image”

Attach an image to the comment that triggers ReviewPhin, or add one to the merge request description. ReviewPhin recognizes standard Markdown images such as ![screenshot](URL) and HTML images such as <img src="URL">. It downloads only images hosted by the configured GitLab instance; links to other websites are not fetched.

Up to 10 referenced images and 25 MiB of image data are downloaded for one run. Each downloaded image also has a 10 MiB limit. Before sending an image, ReviewPhin checks the selected model’s vision limits and resizes or re-encodes the image when necessary. A failed or omitted image does not stop the run: ReviewPhin continues with the available text and images, and download or model-limit omissions are reported as unavailable.

A model must support image input to inspect attachments. When the selected model does not support images, ReviewPhin continues with text and tells the model that the images were omitted.

Override the model for one merge request

Section titled “Override the model for one merge request”

Add a directive to the merge request description (not a comment):

/reviewphin-profile byok-gpt5.4

That named profile wins over the tenant profile, the database default, and the Copilot CLI fallback for every run on this merge request. See model profiles for how profiles are defined and resolved.

Teach a convention

Section titled “Teach a convention”
@reviewphin for future reference, prefer functional React components over class components

Other triggers include remember, please remember, going forward, in the future, team policy, stable preference, always prefer, and please prefer. The convention is stored in project memory and applied to future reviews.

Where reviews arrive

Section titled “Where reviews arrive”

Findings post as bot-owned discussions, one per finding, plus a single summary comment at the top of the discussion list. Replying inside any bot-owned discussion continues that finding. For the shared command list across platforms, see comments and triggers.