portfolio-run
Use when choosing and continuing ready work across multiple toolbox products while preserving one primary scenario per task.
- License 许可
- MIT
- 最近更新
- 2026-07-18
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- portfolio-run
- description
- Use when choosing and continuing ready work across multiple toolbox products while preserving one primary scenario per task.
02 资源文件 1
- 参考文档agents/openai.yaml214 B
03 SKILL.md 正文
Portfolio Run
Advance products serially through deterministic plans. Portfolio selection does not create a multi-scenario task.
Loop
- Run
toolbox workbench check; keepprofile.languagefor operational prose. - Run
toolbox portfolio run-plan. Reportskipped_productswith their exact blocker codes and refs. - Check workbench task status. When no task is active, join the selected
scenario to an existing issue and invoke the
task-startskill for that issue. If no issue exists, useticket-incubateto propose one and stop at its human approval gate. Never invent an issue ID or create task state in toolbox. - Invoke
product-runfor exactly one selectedscenario_ref. Never combine a selected scenario withremaining_candidatesin one task. - If that product reaches a policy, evidence, acceptance, or human gate, record the exact task blocker. Continue only when the product can be left without claiming completion or acceptance.
- Choose the first
remaining_candidatesentry whose product was not attempted in this session, validate it withtoolbox product run-plan <product-id> --scenario <scenario-id>, and invokeproduct-runagain. - Re-evaluate with
toolbox portfolio run-planafter any persisted scenario or product transition. Stop when no ready candidate remains or a portfolio-wide integrity failure occurs.
Boundaries
- Keep refs, state values, and reason codes in canonical English; explain them in
profile.language. - Do not convert an unresolved human or policy gate into a completed scenario.
- Keep one scenario per task; start a new task before continuing another product.
- Do not schedule parallel writers, auto-merge PRs, or deploy production.
04 同仓库更多 Skills
ticket-incubate
Use to turn an idea that has no issue/ticket yet into an actionable GitHub issue. Grow a vague notion by divergence, narrow only the gaps via a deficit-based interview, and create the issue once the four required elements (background, goal, done criteria, out-of-scope) are filled. Triggers on "I have an idea", "let's make this a ticket", "create an issue", "interview me", or describing something to make/fix/improve without an issue number, even if the user never says the word "ticket". It is the front of the workbench lifecycle (idea -> ticket-incubate -> issue -> task-start). Distinct from brainstorming/design skills: those work out the how of what to build now; this captures the what of later work as a well-honed ticket.
task-tickets
Use to scan open workbench tickets joined with task branches and lifecycle events into unstarted / claim-attempted / in-progress / awaiting-PR / cleaned-up states. Read-only.
task-submit
Use when the user asks to submit, wrap up, or extract from a task, or when work has reached a point to open a PR.
task-status
Use to see local and remote task status at a glance and diagnose drift (unpushed, stale status, missed cleanup). When you want to know where everything stands before starting or submitting, or after a multi-device handoff.
task-start
Use when the user asks to start a task, pick up an issue, or begin a new attempt, or hands over an issue number to start work.
task-done
Use when the user asks to clean up or discard a task, or to clear the workspace after a PR is merged.