An engineer opens a merge request by hand: a chart bump plus one
extra setting. Someone types /review all. Several
reviewer personas read it in parallel, each backed by a different
model and each saying which one. A final comment tallies the verdicts
and pulls out the one finding only one of them caught.
/review comment.
Here an engineer bumps ingress-nginx by a minor and,
in the same MR, turns on allowSnippetAnnotations so one
team can add a custom header. Two lines, easy to approve on a
Friday.
Putting the assistant in the MR's reviewers list runs the
default persona once per commit. A /review comment
does the same; /review all runs every configured
persona; /review premium runs just one.
Only people on the trigger allowlist can start a review by comment. Unknown persona names are dropped with a warning and the rest still run.
Personas are the interface your team learns: default for a balanced review, premium for high-stakes MRs, quick for a cheap second opinion. Each maps to a model profile in configuration. When a better model ships, the pointer changes and nothing else does.
Reviews are keyed on MR, commit and persona, so a persona never reviews the same commit twice. A force-push makes it eligible again.
Each reviewer sees the diff and the MR description, and can be given read-only tools to open neighbouring files: the values the chart already sets, the Ingress objects that would use the new setting In development.
An approval that needed evidence it did not have (a changelog, a CRD schema) is marked contested rather than passed through as a clean yes.
You always know which model produced which opinion. Each comment carries a verdict, a quality score, what was verified and what worries it, and the commit it reviewed.
Here two personas approve. premium requests changes: the snippet setting lets anyone who can write an Ingress inject NGINX configuration, which is why the chart turned it off by default.
When two or more personas ran, a synthesis comment tallies the verdicts, lists shared concerns and highlights unique findings. A lone dissent is not averaged away; it is the headline.
The MR is stamped review::LGTM, review::NC
or review::COMMENT, taking the most cautious persona's
verdict. A re-review replaces the label, so the MR list always
shows the current state.
The assistant's own reviewer shares its prompts and its blind spots. So every MR the assistant opens is also read cold by an independent reviewer that knows nothing about how the MR was produced. It checks that the title matches the diff, that nothing secret leaked, and that the change lands somewhere that is actually deployed.
Its verdict is a separate label: xrev::LGTM,
xrev::NC (needs changes) or xrev::DNM (do
not merge).
infrastructure/ingress/helmrelease_ingress-nginx.yaml chart: ingress-nginx - version: 4.10.1 + version: 4.11.2 values: controller: + allowSnippetAnnotations: true replicaCount: 3
/review all→ read_repo_file(infrastructure/ingress/helmrelease_ingress-nginx.yaml) controller.config: none · no snippet allow-list → list_repo_dir(apps/) 31 Ingress objects across 9 namespaces → read_repo_file(apps/checkout/ingress.yaml) configuration-snippet: more_set_headers "X-Frame…"
default — model: qwen3.5-397b — approvepremium — model: claude-opus-4-7 — request_changescontroller.addHeaders entry or scope it with an admission policy.quick — model: claude-haiku-4-5 — approve- allowSnippetAnnotations: true + addHeaders: { X-Frame-Options: DENY } /review premium → approve · label replaced: review::LGTM
Illustrative. The comment headers, the synthesis and the label scopes follow the format the reviewer writes; the model names are examples and are set per installation.
sam/ingress-nginx-4.11 into main · 1 file · +2 −1 · reviewers: priya, gitops-assistantpremium — model: claude-opus-4-7 — request_changesQuality: 91/100
The version bump is fine on its own. The second change is not scoped to what the MR description asks for.
replicaCount and existing values preserved byte-identicalcontroller.allowSnippetAnnotations: true applies to every Ingress in the cluster (31 objects, 9 namespaces). Any team that can write an Ingress can then inject NGINX configuration. The chart defaults this to off for that reason.checkout. controller.addHeaders (or a per-host header ConfigMap) covers it without enabling snippets.Generated by the GitOps Assistant reviewer · head_sha a41c07e
default — model: qwen3.5-397b — approveQuality: 84/100
Generated by the GitOps Assistant reviewer · head_sha a41c07e
quick — model: claude-haiku-4-5 — approveQuality: 78/100
LGTM: minor chart bump, one values addition.
Generated by the GitOps Assistant reviewer · head_sha a41c07e
default, premium, quick2/3 personas approve; premium requests changes. Shared concern: none that all three raised. default noted the snippet setting but did not block on it.
Unique finding: premium showed that the snippet setting is cluster-wide while the need is one header for one service, and named a narrower alternative. The operator should resolve this before merging.
quick's approval is marked contested: it lacked changelog evidence.
Synthesis produced by the GitOps Assistant reviewer. Read alongside the individual persona comments.
Fair, switched to addHeaders. /review premium
premium — model: claude-opus-4-7 — approve head_sha b92e310Snippets stay disabled; the header is set globally by the controller, which matches the stated need. Label updated to review::LGTM.
A reviewer that agrees with everything is worse than none. These are the rules that keep the verdicts meaningful.
/review is ignored.xrev::LGTM / NC / DNM labels/review comment (built, not yet enabled in production)review:: verdict labels and contested-approval markingreview <MR> from Slackreviewer:all