English · 简体中文
A record of my Spec Kit workflow: its configuration, skills, constitution, and how they evolve over time.
Report a bug · Request a feature
Table of Contents
A record of my Spec Kit workflow: its configuration, skills, constitution, and how they evolve over time.
See the open issues for planned features and known issues.
- Spec Kit, verified with version 1.0.6 (pinned in
.specify/init-options.json), installed in your project and providing the commands the workflow prompt calls:/speckit-plan,/speckit-tasks,/speckit-analyze,/speckit-implementand/speckit-converge. - Claude Code to run the prompt's leading
/goalcommand. The workflow is verified with Claude Code; other agents that support/goaland the Spec Kit commands may work but are unverified. - An existing feature spec, written with
/speckit-specify(see Usage).
Use the workflow prompt only after specification work is finished: the feature spec has been
written with /speckit-specify and, if you use it, clarified with /speckit-clarify.
The whole workflow runs in four steps:
/speckit-constitution— set the project's principles./speckit-specify— write the feature spec./speckit-clarify— optional: resolve open questions in the spec.- Paste the workflow prompt below — it runs plan, tasks, analyze, implement and converge in a loop until the feature is done.
What the prompt does. It runs /speckit-plan and /speckit-tasks, skipping both when a plan
and tasks already exist. It then runs /speckit-analyze and fixes every medium or higher issue in
the plan and tasks, up to 20 fix rounds, without ever editing the spec; if an issue comes from the
spec itself, it stops and asks you to fix the spec. Next it runs /speckit-implement on the
unfinished tasks, continuing past individual failures, and /speckit-converge to find anything
still missing; each gap goes back into the tasks and the loop repeats, up to 10 gap rounds. You
can rerun the prompt after changing the spec.
What you get back. A status summary of completed tasks, unfinished tasks and spec-related issues. When a retry limit is reached, or the spec needs manual changes, a report message comes before the summary.
The prompt is intentionally kept in its original Chinese, exactly as the author uses it. Copy it with the copy button on the block:
Workflow version: v1.0.0
/goal 按照顺序串行执行。
0. 如果plan和tasks已存在（例如修改spec后重跑），跳转到3
1. /speckit-plan
2. /speckit-tasks
3. /speckit-analyze
4. 确认是否有medium及以上的问题（不满足第2步要求的任务也视为medium问题）。如果有问题的根源在spec（如需求模糊、冲突或缺失，需指出spec的具体条目），跳转到10，并携带“以下问题需要人工修改spec”的报告消息（附问题清单）；如果只有其他问题，跳转到5；无问题跳转到6
5. 如果本步骤已执行满20次（每次进入第6步时计数清零），跳转到10，并携带“为什么对计划反复修复了20次都无法完整修复”的报告消息；否则修复medium及以上的问题（只修改plan和tasks，使其与spec保持一致，不得修改spec，不得改变任务的完成状态），然后跳转到3
6. /speckit-implement 只执行未完成的任务，单个任务失败不中断，失败任务保持未完成状态，交由后续converge处理。
7. /speckit-converge
8. 确认是否有缺口，如果有缺口跳转到9，无缺口跳转到10
9. 如果本步骤已执行满10次，跳转到10，并携带“为什么反复修复了10次缺口后仍然有缺口”的报告消息；否则确认每个缺口都已写入tasks（新增任务或将相关任务重新标记为未完成，并记录上一轮失败的原因），然后跳转到3
10. 输出状态摘要（已完成任务、未完成任务、spec相关问题）；有报告消息则附在摘要之前
11. Done，结束goal
Contributions are welcome. Read CONTRIBUTING.md for how to open an issue or a pull request, and CODE_OF_CONDUCT.md for the standards expected of everyone taking part.
Please do not report security issues in public issues or pull requests. SECURITY.md explains how to report them privately.
Distributed under the MIT License. See LICENSE for details.
Project link: https://github.com/anyingiit/my-speckit-workflow