board-column-flow
Use when moving tasks or reporting status - the kanban column semantics, which columns are human/architect/QA gates, and where the PM actually acts
SKILL.mdcatalog/agents/product-manager/skills/board-column-flow/SKILL.md
GitHubBoard Column Flow
Overview
The board has more columns than the PM acts in. Knowing which columns belong to the human, the architect, and QA keeps you from moving a task through a gate that isn't yours.
The columns
Two spines — a given task only passes through the columns its type needs:
task / bug: backlog → todo → in_progress → code_review → ready_for_qa
→ in_qa → pm_uat → human_uat → done → released
analiz: backlog → todo → in_progress → analiz_review → done → released
need_revision ← any gate that rejects (code_review, in_qa, pm_uat, human_uat, analiz_review)
→ back to in_progress' owner, then forward again from where it left
blocked ← parked here automatically while any blocked_by task is open
→ released automatically when the last one lands
task_type: "technical" (no user-facing behaviour) skips pm_uat — QA routes it straight to human_uat. need_revision is not a stage in the line — it is where every gate sends work back, and the PM is not dispatched there: act on a need_revision task only if the stakeholder raises it in chat. done → released is release-engineer's step, not yours.
Who owns each gate
| Column | Owner | PM action? |
|---|---|---|
| backlog | PM | Yes — create tasks here by default; stakeholder prioritizes |
| todo | assignee agent (auto) | Move approved tasks here to start them |
| in_progress | developer | No |
| analiz_review | human | No — the human approves the architect's plan (→ done) or rejects it (→ need_revision) |
| code_review | system-architect | No — architect advances to ready_for_qa or need_revision |
| ready_for_qa | QA (queue — QA takes it into in_qa) | No |
| in_qa | QA (testing in progress) | No |
| need_revision | developer / architect | Not dispatched here; act only if the stakeholder asks in chat |
| pm_uat | PM | Yes — verify against AC by walking each flow yourself in the browser or on a device; QA's evidence is a cross-check, not a substitute (see pm-uat-review). Skipped for task_type: "technical". |
| human_uat | stakeholder | No — stakeholder reviews |
| done / released | — | Terminal; released per project convention, by release-engineer |
PM action points (only these)
- backlog: create tasks here; the stakeholder reviews/prioritizes before agents pick them up.
- todo: move approved tasks here to trigger the assignee.
- pm_uat: review against AC by walking each one yourself in the browser or on a device (QA's evidence is a cross-check, not a substitute). All AC confirmed → approve each with
review_criterionand move tohuman_uat— no comment. Any gap → reject those criteria viareview_criterion, a numbered gap list as comment, move toneed_revision.
Everything else (analiz_review, code_review, QA columns, human_uat) is another actor's gate — don't move tasks through them.
Common Mistakes
- Moving an analiz task out of
analiz_review— that's the human's approval gate. - Advancing a task in
code_review— that's the architect's. - Approving in
pm_uatby reading code, or on QA's evidence alone, instead of walking the flow yourself. - Acting on a
need_revisiontask uninvited — you are not dispatched there.
Red Flags
- You moved a task through a column not in the PM action-points list.
- A
pm_uatapproval whosereview_criterionnote cites nothing you observed in this run.
Board mechanics
- Claiming a task and moving it between columns takes seconds and announces what you are doing; it produces nothing by itself. Do it inside the step that does the work, never a step of its own and never as the first item of a plan — a step whose only content is a claim or a move is rejected before it runs.
- The task is already in the column named in your context; never plan a move into the column it is already in.
- If the payload says resumed=question_answered: you previously stopped on the question in payload.question and the human replied in payload.answer — continue from where you stopped using that answer; do not ask it again.