Context
Epic #4189 (AI-generated E2E test coverage) is code-complete — 11 of its 12 sub-issues are shipped;
only #4199 (enable + validate end-to-end) remains open, and it's a rollout/observation task, not
engineering.
One real gap surfaced while reviewing the epic for #4199 readiness: the @gittensory generate-tests
command and the manifest_missing_tests auto-trigger both exist and work, but nothing in ORB's own
review comment tells a maintainer the command exists. manifest_missing_tests already renders as a
finding in the unified comment (publicSafeManifestPolicyFinding, src/queue/processors.ts:1031) with
action text ("Add regression/invariant coverage, update relevant tests, or attach passing validation
output...") that never mentions the AI-generation option — a maintainer would have to already know the
slash command from documentation.
CodeRabbit's equivalent feature avoids this exact gap: their "Generate unit tests" affordance is a
checkbox surfaced directly inside their own PR walkthrough comment, not a separate dashboard toggle —
so the option is discoverable at the point a maintainer is already looking (docs.coderabbit.ai/finishing-touches/unit-test-generation).
Deliverable
When manifest_missing_tests fires and the e2eTests converged feature is enabled for the repo
(resolveConvergedFeature(env, manifest, "e2eTests", repoFullName) — same check the #4196 auto-trigger
already uses immediately below this code), append a short, discoverable mention of
@gittensory generate-tests to that finding's rendered action text in the main review comment.
- No new UI surface, no dashboard — text-only, inline in the comment ORB already posts (matches the
maintainer's explicit preference: this should live in the review comment, not the dashboard).
- Gated on the SAME feature flag the auto-trigger already checks, so an unconfigured repo is never
told about a command that would just bounce with "not enabled."
- Byte-identical output for every repo that doesn't have
e2eTests on — this is additive text only,
not a new finding/code.
Expected outcome
A maintainer whose PR trips manifest_missing_tests sees, right in ORB's own comment, that they can
comment @gittensory generate-tests to get an AI-generated Playwright test for the PR — the same
"discoverable at the point of use" property CodeRabbit's walkthrough checkbox has, without requiring
a new dashboard settings surface.
Related: #4189 (epic), #4196 (the auto-trigger this reuses the exact same feature-gate check from),
#4199 (the still-open enable+validate step this directly improves the readiness of).
Context
Epic #4189 (AI-generated E2E test coverage) is code-complete — 11 of its 12 sub-issues are shipped;
only #4199 (enable + validate end-to-end) remains open, and it's a rollout/observation task, not
engineering.
One real gap surfaced while reviewing the epic for #4199 readiness: the
@gittensory generate-testscommand and the
manifest_missing_testsauto-trigger both exist and work, but nothing in ORB's ownreview comment tells a maintainer the command exists.
manifest_missing_testsalready renders as afinding in the unified comment (
publicSafeManifestPolicyFinding, src/queue/processors.ts:1031) withaction text ("Add regression/invariant coverage, update relevant tests, or attach passing validation
output...") that never mentions the AI-generation option — a maintainer would have to already know the
slash command from documentation.
CodeRabbit's equivalent feature avoids this exact gap: their "Generate unit tests" affordance is a
checkbox surfaced directly inside their own PR walkthrough comment, not a separate dashboard toggle —
so the option is discoverable at the point a maintainer is already looking (docs.coderabbit.ai/finishing-touches/unit-test-generation).
Deliverable
When
manifest_missing_testsfires and thee2eTestsconverged feature is enabled for the repo(
resolveConvergedFeature(env, manifest, "e2eTests", repoFullName)— same check the #4196 auto-triggeralready uses immediately below this code), append a short, discoverable mention of
@gittensory generate-teststo that finding's rendered action text in the main review comment.maintainer's explicit preference: this should live in the review comment, not the dashboard).
told about a command that would just bounce with "not enabled."
e2eTestson — this is additive text only,not a new finding/code.
Expected outcome
A maintainer whose PR trips
manifest_missing_testssees, right in ORB's own comment, that they cancomment
@gittensory generate-teststo get an AI-generated Playwright test for the PR — the same"discoverable at the point of use" property CodeRabbit's walkthrough checkbox has, without requiring
a new dashboard settings surface.
Related: #4189 (epic), #4196 (the auto-trigger this reuses the exact same feature-gate check from),
#4199 (the still-open enable+validate step this directly improves the readiness of).