菜单
Run a full orchestrated iteration — PM decomposes the goal, delegates to specialist roles, collects outputs, and synthesizes results
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
One-command project setup for OpenUP - interactive initialization wizard
Mark a task as complete, update roadmap, commit changes, and prepare traceability logs
Execute an already-claimed lane's Operations boxes with script/judgment classification — script steps run directly with zero self-brief, judgment steps self-brief and execute. Handles only pick/resume; every other resolve path routes to /openup-next.
Run OpenUP Assess Results at iteration end — check evaluation criteria, demo only completed acceptance-tested work, feed discovered work back, and trigger the milestone review at a phase boundary
Run the phase milestone go/no-go — prepare derived evidence, pause for the human decision via an input-request, and record the milestone (never advance the phase itself)
Begin a new OpenUP iteration with proper phase context and task selection
| name | openup-orchestrate |
| description | Run a full orchestrated iteration — PM decomposes the goal, delegates to specialist roles, collects outputs, and synthesizes results |
| model | inherit |
| fit | {"great":["complex multi-role tasks","architecture+impl+test cycles","anything benefiting from role isolation"],"ok":["medium tasks where decomposition aids token-efficiency"],"poor":["trivial changes","single-role work","hotfixes — heavyweight overkill"]} |
| arguments | [{"name":"task_id","description":"The task ID to orchestrate (must match a task in docs/roadmap.md)","required":true},{"name":"team","description":"Team type to use (feature, construction, elaboration, inception, transition, investigation, planning, full). Auto-selected from phase if not provided.","required":false},{"name":"dry_run","description":"Preview the orchestration plan without spawning teammates (true/false, default: false)","required":false}] |
The Project Manager orchestrates a full iteration by decomposing the goal into role-specific subtasks, delegating to specialist teammates, collecting their outputs, and synthesizing a coherent result.
This skill implements the coordinator + specialist pattern: the PM acts as an orchestrator (coordinator) that spawns focused agents (specialists) with isolated task contexts, then integrates their outputs.
/openup-start-iterationgit rev-parse --abbrev-ref HEAD returns a non-trunk branch)docs/iteration-plan.md must exist with the iteration goal and acceptance criteriaRead the following (in order):
docs/iteration-plan.md — iteration goal, acceptance criteria, task listdocs/project-status.md — current phase, active taskdocs/roadmap.md — task details for $ARGUMENTS[task_id].claude/memory/iteration-learnings.md — past learnings to avoid repeating mistakesAnalyze the iteration goal and break it down by role. For each role in the active team, determine:
Document the decomposition before spawning any teammates:
Orchestration Plan for [task_id]:
Analyst: [subtask] → Deliverable: [output] → Done when: [criteria]
Architect: [subtask] → Deliverable: [output] → Done when: [criteria]
Developer: [subtask] → Deliverable: [output] → Done when: [criteria]
Tester: [subtask] → Deliverable: [output] → Done when: [criteria]
If $ARGUMENTS[dry_run] is true: output this plan and stop. Do not spawn teammates.
If team is not already active, determine the team type:
$ARGUMENTS[team] if provided/openup-start-iteration step 7)Spawn only the roles needed for this iteration's decomposition. Do not spawn roles that have no meaningful contribution to this specific task.
Send each teammate a focused brief using the PM's delegation format:
[ROLE]: Your task for this iteration is: [focused scope].
Context you need:
- [specific doc 1 with path]
- [specific doc 2 with path]
- [constraint or decision that affects their work]
Deliverable: [specific output — section of a document, code changes, test results]
Done when:
- [criterion 1]
- [criterion 2]
- [criterion 3]
Key rule: Give each specialist only the context they need. Do not dump the full project into every brief.
As each specialist completes their work, collect:
Check each output against the "done when" criteria from their brief. If criteria are not met, send a follow-up brief with specific gaps to address.
Integrate all specialist outputs:
docs/iteration-plan.md/openup-assess-completeness scope: iteration to get a final rubric-based assessmentIf the assessment returns needs_revision: identify which specialist needs to address which gaps, and send targeted follow-up briefs (back to step 4 for the relevant role).
Append to .claude/memory/iteration-learnings.md:
## [YYYY-MM-DD] [task_id]: [task title] — Orchestrated
- **Team composition**: [roles used and why]
- **Delegation that worked well**: [what made a particular brief effective]
- **Coordination overhead**: [where handoffs created friction and how to avoid next time]
- **Synthesis challenges**: [conflicts between specialist outputs and how resolved]
Once all acceptance criteria are met and the rubric assessment is satisfied, invoke /openup-complete-task task_id: $ARGUMENTS[task_id] to commit, log, and create the PR.
Returns: