pm-as-solo-orchestrator
Use when planning a delivery request as the only orchestration agent - one board-writing subtask, backlog first, one approval question, then move approved tasks to todo
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 122
- Forks
- 18
01 前言 Front Matter
- name
- pm-as-solo-orchestrator
- category
- pm
- description
- Use when planning a delivery request as the only orchestration agent - one board-writing subtask, backlog first, one approval question, then move approved tasks to todo
03 SKILL.md 正文
PM as Solo Orchestrator
Overview
When product-manager is the only enabled orchestration agent, the plan you write for a request is not a multi-agent handoff — it's one subtask that writes the board, followed by columns doing the rest of the work automatically.
Core principle: One subtask creates the tasks; the board, not the plan, drives every stage after that.
Rules
- All subtasks are assigned to product-manager — never to system-architect, security-agent, backend-developer, frontend-developer, mobile-developer, data-scientist, game-developer, or qa-agent. Developer work goes through
create_board_task, not an orchestration subtask. - Multiple parallel PM subtasks are fine only when every one is read-only (e.g. web research +
list_board_tasksconcurrently). At most one subtask per parallel group may write to the board; a second board writer in the same group rejects the whole plan. Usedepends_onto sequence when a later subtask needs an earlier one's result. - Delivery stages (analiz, implementation, QA verification, approval) are board columns the task travels through afterwards — never plan a subtask per stage; that opens one board record per stage for a single piece of work and is rejected before the plan runs.
- Analiz tasks skip approval — they go straight to
todo(investigation is always safe to start). - Planner sets
ready=falseonly when a product decision blocks creating any board task (max 3 questions, see stakeholder-intake's ask-vs-assume rule). Otherwiseready=true.
Standard flow — one subtask, in order
- Create the tasks in
backlog(see implementation-task-spec, acceptance-criteria-gwt). - Call
ask_user, choice mode:context: what you asked / what you're assuming (name each assumption so the stakeholder can correct it cheaply) / the task list with assignee and order.- options: "Start all (recommended)", "Start only some", "Change something", plus whatever else applies.
- On approval, move every approved task to
todowithmove_board_task. Order lives inblocked_by— never hold a task back inbacklogto fake an order; a blocked task parks itself and starts automatically once its blocker lands. - On partial approval, move only the selected tasks; leave the rest in
backlog. - On rejection or requested changes,
update_board_taskthe affected fields, ordelete_board_task+create_board_taskforassignee/derived_from(creation-only fields) — and fortask_typewhenupdate_board_task's schema does not list it. - Report which tasks moved to
todo(agents will pick them up) and which stay inbacklog, and why.
This is ONE subtask calling these tools in order — not three subtasks. Splitting it hands the same board record to several agents that cannot see each other's writes.
If a blocking product decision remains, call ask_user within a subtask — never append questions as markdown in the final summary.
Common Mistakes
- A subtask per delivery stage instead of one subtask that creates the tasks.
- Holding implementation tasks in
backlog"until analiz is done" instead of usingblocked_by— the board already enforces that order. - Two parallel subtasks writing to the board in the same group.
- Approval-question
contextthat doesn't separate what you asked from what you assumed.
Red Flags
- A plan with a subtask named "QA stage" or "human_uat review" — those are columns, not subtasks.
- More than one board-writing subtask in a parallel group.
- A task left in
backlogwith a "Depends on: …" sentence instead ofblocked_by.
04 同仓库更多 Skills
visual-craft
暂无简介
ux-copy
Use when you write any words a user reads in a mockup or a hand-off - content language and voice, buttons, labels, errors, empty states, confirmations, numbers and dates, and the longest-string check
screen-mockup-html
Use when you draw screens as the design task's HTML mockup - ONE canvas document holding every screen and every variant side by side, the sanitizer-safe recipe, phone and wide frames, the states grid, static control states, and the loopback self-check with screenshots at 375, 768 and 1440 attached to the task
performance-awareness
How the performance score moves for your role. Use when the score context shows a declining trend, or before handing off work you have not self-checked against every criterion.
design-variants
Use when a screen design needs variants - how many, variants as side-by-side rows on the one canvas, what makes two variants genuinely different, recommending one, how the human's choice is read, and narrowing to the chosen one in revision
design-system-authoring
Use when you derive, change or extend a design system - where tokens live per stack, the DTCG token tree, role-based names for both themes, base vs layer vs drift, the DESIGN.md section order, the inventory format, the contrast check and the lint every proposal returns