Agentify · GitLab workflow

the flow

Six moves from issue to draft MR.

Each step is a skill in the bundle. GitLab projects keep their own label taxonomy; the workflow prefers existing labels and only touches issues you have opted in.

1triage the board

gitlab-triage

Runs issues through a conservative state machine — needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix — preferring existing project labels over inventing new ones. Confirms the repo with glab repo view before any mutation. Opting in means ready-for-agent (plus agentify-ready for fan-out). New issues are shaped first: the grill-me skill interviews you one question at a time until the requirement is concrete, then files it with glab issue create.

2pick up an item

glab-autopilot

Resolves the GitLab context with glab: confirms glab auth status and glab repo view, selects the target issue or MR (explicit number, URL, latest, or first), and turns glab issue view into an execution checklist.

3implement in isolation

worktree-autopilot

Creates a fresh branch and git worktree off the correct base, implements the smallest viable change there, and verifies it with the repo's own formatting, linting, type checks, and tests. Commits with a conventional message and leaves the original checkout untouched.

4fan out in parallel

issue-killer

When several issues are opted in, launches one tmux pane and one worktree per issue via glab issue list --label agentify-ready, each running an agent that opens a draft MR. Supervise them all with tmux attach -t issue-killer.

5raise a draft MR

pr-creator

Validates glab and auth, confirms base/source branch, title, and body, then creates a draft merge request with glab mr create --draft and returns the link. Never opens a non-draft MR unless you explicitly ask.

6carry context across sessions

agentify ctx

Fan-outs and handoffs are recorded with agentify ctx note so a later session knows what is in flight. Every worktree gets its own .agentify/ store, so notes and tracking stay per-checkout.

bundle

What workflow install glab lays down.

Seven skills, installed together and templated for your provider.

gitlab-triageConservative label/state management for issues.

grill-meInterviews you until a new item is concrete, then files it with glab issue create.

glab-autopilotResolves GitLab issue/MR context with glab.

issue-killerParallel tmux panes + worktrees for opted-in issues.

worktree-autopilotIsolated implement, verify, commit.

pr-creatorDraft MR creation with prerequisite checks.

commit-creatorFocused, Conventional-Commit staging and messages.

platform cli

Readiness and board queries.

The workflow confirms the toolchain and the repo before it touches anything on GitLab.

glab readiness · board

# readiness
glab auth status
glab repo view

# the opted-in board
glab issue list --label agentify-ready --output json | head -c 20000

# load one issue as a checklist
glab issue view <number> --output json

guardrails

Conservative by default.

opt-in only

Issues are selected by an opt-in label or explicit URLs/IDs — never arbitrary open issues. One worktree per issue, never shared.

draft MRs

Agents open draft merge requests by default (glab mr create --draft). Nothing is marked ready for review automatically.

no force / merge

No force-push, no rewriting shared history, and no closing issues or merging MRs unless you explicitly ask. Labels are never created without approval.

ai disclaimer

Every triage comment posted to GitLab carries > This was generated by AI during triage. at the top of the body.

From board to draft MR — the agent drives the whole loop.