GitLab Board Sync

Mirror local SDD planning/execution state onto existing GitLab issues. Does not create issues.

Prerequisites

1. GitLab MCP available — follow mcp-gitlab-usage for all tool calls
2. If multiple GitLab MCP servers: ask human which to use
3. docs/context/gitlab-sync-config.md validated
4. For post-planning batch: gitlab-issue-feature-map.md exists

Configuration

Read references/gitlab-sync-config.template.md for expected fields:

• projectid, milestoneid, assignee_default
• statusbacklog, statusinprogress, statusdone, status_blocked
• rflabelformat, basebranch, workbranch, mrtargetbranch

Discover unknown status labels via listprojectlabels — confirm with human.

Flow C — Pre-implementation read

When task lists GitLab issue IIDs:

1. read_issue for description + recent comments
2. Non-blocking if inaccessible — log warning, continue
3. Do not mutate issue in this flow

Flow A — Post-planning batch

Per issue in map (excluding exclude_issues):

1. Validate milestone (project + group)
2. updateissue with milestoneid
3. One setissuelabels — removelabels + addlabels atomically (RF, Status, Equipe)
4. assign_issue
5. setissueestimate from linked task header (seconds) only if timestats.timeestimate is empty — never overwrite; skip values < 60

Flow B — Per-task execution sync (legacy 1:1)

Mutual exclusion: any issue_iid in delivery-units.md → do not run Flow B. Owner = ../ns-spec-driven/references/delivery-units.md GitLab status/spent (SSoT).

When task row maps 1:1 to its own GitLab issue (no delivery units file, or unpublished units + legacy gitlab-issue-feature-map.md):

Task start (before coding)

remove: status_backlog
add: status_in_progress

Record STARTTIME / STARTEPOCH when coding starts (for wall-clock addissuespent_time — see ../ns-execution-gitlab-issue/references/time-tracking.md).

Task complete (after validation)

Only if already statusinprogress:

remove: status_in_progress
add: status_done, RF: NNN
add_issue_spent_time
add_issue_comment (internal=true)

Never backlog → done in one step.

Flow D — Per delivery unit

Run only when SSoT row is Flow D (published issue_iid + local execute, no G). When G owns lifecycle → skip entire Flow D.

Unit start (before coding unit tasks)

remove: status_backlog
add: status_in_progress

Record STARTTIME / STARTEPOCH when coding starts for unit (wall-clock for unit-level spent).

Unit complete (after unit validation + review approved)

Only if already statusinprogress:

remove: status_in_progress
add: status_done
add_issue_comment (internal=true)

Local-only (SSoT Flow D row): include addissuespenttime once per unit (wall-clock, not estimatesum); set spent_posted = yes.

Never backlog → done in one step. Never addissuespent_time per task inside unit. Never Flow D when G owns the same issue.

Anti-patterns

References