plane × gitlab tickets: the feasibility verdict
Status: open design. The refusal stands: there is no GitLab work-item plug on this Plane pin. Ruling (2026-08-22):
gitlabon the plane app should mean WORK-ITEM integration with the estate’s GitLab (git.corp on the corp estate), with SSO belonging tooidc.plug.yamland the GitLab-OAuth chain (plane-set-god-mode-gitlab.job.yaml) as the Community interim. This page records why that plug cannot exist against the v1.4.1 pin, what would unlock it, and what Community can honestly express.
Everything below was measured against the makeplane/plane v1.4.1 tag
(the pinned images’ source) on 2026-08-22, not inferred from a feature matrix.
1. What Community v1.4.1 actually ships for “gitlab”
Section titled “1. What Community v1.4.1 actually ships for “gitlab””Login, and only login. The closed instance-configuration key list
(apps/api/plane/utils/instance_config_variables/core.py:101-127) carries
exactly five gitlab keys: IS_GITLAB_ENABLED, GITLAB_HOST,
GITLAB_CLIENT_ID, GITLAB_CLIENT_SECRET, ENABLE_GITLAB_SYNC. The last is
not work-item sync: authentication/adapter/base.py (check_sync_enabled,
lines 125-137) gates avatar/profile refresh AT OAUTH LOGIN. The God Mode
surface (PATCH /api/instances/configurations/) bulk-updates only EXISTING
rows and ignores unknown keys without error, so there is no key a tickets
plug could set and no error it would even get for trying.
No integration service. The work-item integration (bidirectional issue
sync, MR-state automation) is Plane’s Silo service. Community’s repo at
v1.4.1 has no silo app (docker-compose.yml services: web, admin, space,
api, worker, beat-worker, migrator, live, db, redis, mq, minio, proxy) and no
integration code. What it has is silo-AWARE residue that proves the split:
silo in the reserved workspace-slug list
(packages/constants/src/workspace.ts:34,
apps/api/plane/utils/constants.py:27) and as the example value for the
WEBHOOK_ALLOWED_HOSTS SSRF allowlist (plane/settings/common.py:74).
No importers. plane/db/models/importer.py is a routeless legacy model;
the two name="importer" routes in plane/app/urls/external.py are the GPT
endpoints. The old Jira/GitHub importers are gone from Community.
2. What unlocks it
Section titled “2. What unlocks it”The GitLab integration is priced Pro/Business/Enterprise
(plane.so/marketplace/gitlab; docs.plane.so/integrations/gitlab is badged
Pro). Self-hosted it ships only with the Commercial edition, the paid
self-host whose deploy includes Silo; the integration’s OAuth callbacks land
on /silo/api/gitlab/...
(developers.plane.so/self-hosting/govern/integrations/gitlab). Community has
Free-tier parity by design, so no future Community minor should be expected
to grow this; re-measure instance_config_variables/core.py at each pin
bump anyway (the God Mode gitlab job already preflights the key list at
runtime). One honesty note: even on paid, GitLab lags GitHub; upstream
issue makeplane/plane#8088 (open, assigned) asks for issue-sync parity;
the marketplace page leads with MR linking.
3. What Community CAN express: and why it is still not a plug
Section titled “3. What Community CAN express: and why it is still not a plug”Community ships two real doors, both measured on the v1.4.1 source:
- v1 REST API behind
X-Api-Keypersonal tokens (plane/api/): issue links (IssueLinkListCreateAPIEndpoint,views/issue.py:1113,IssueLinkis title + url + free JSON metadata) and comments (views/issue.py:1360). A GitLab CI job or a webhook bridge can attach MR/issue URLs and comments to work items today. - Outbound workspace webhooks (
plane/app/urls/webhook.py), for the Plane→GitLab direction.
But Plane Community has NO inbound GitLab receiver, so anything beyond
“CI curls the Plane API” needs a translator listening for GitLab webhooks
and speaking Plane’s API, new SOFTWARE, not configuration. The catalogue’s
precedent is exact: apps/standalone/gitlab-huly is that translator for
Huly, and it is an APP with its own service, never a plug on the tracker.
A plane plug consuming the gitlab socket would have nothing to converge:
the socket mints tokens and provisions webhooks (both pointing at services
that must exist), and no Plane key or endpoint receives them. That is the
false-success shape (sonarqube’s gitlab.plug.yaml header names it) this
refusal exists to avoid.
If the estate later wants the CE bridge, it is a plane-gitlab-bridge app
in the gitlab-huly mold (GitLab webhook in, Plane v1 API out, plug on the
gitlab socket for its OWN credentials), proposed here as the shape, not
scheduled.
4. Naming: recorded recommendation, deliberately not executed
Section titled “4. Naming: recorded recommendation, deliberately not executed”Live deployments persist plug names and vars: corp’s plane carries
vars.gitlab.host/clientId/clientSecret (the SSO chain), and
tests/unit/catalog/plane_gitlab_test.go pins those spellings. A rename
would orphan the persisted values on the corp deployment for zero
mechanical gain, so the recommendation is:
- Keep
vars.gitlab.*as-is: it is login config and stays spelled the way the deployment already stores it. Its descriptions (and the block comment above it in app.yaml) now say LOGIN out loud. - Reserve the bind name
gitlab(-i gitlab=<stack>) on this app for the future tickets integration; the SSO chain never takes it (it binds no socket today, by design; see the job’s header). - If a tickets surface ever lands, it arrives as
gitlab.plug.yaml(tickets) beside the untouchedvars.gitlab.*(login): plug data and vars are separate namespaces, so the two coexist without migration.
Owner sign-off owed before any of the three moves from “recorded” to “done”.