[Feature]: Add example “hands-off” Spec Kit orchestration workflow for autonomous spec creation
- 主要语言
- Python
- 星标
- 137k
- 派生
- 12.3k
- 平均合并
- 2 天 12 小时
- 30 天内合并 PR
- 159
描述
### Problem Statement
## Summary
I'd like to contribute a repo-hosted example of a "hands-off" Spec Kit orchestration agent/workflow for autonomous **spec creation first**, with safe gates before issue creation or implementation.
The goal is not to change Spec Kit’s core philosophy, but to provide a practical example/pattern for users who want a more automated multi-step flow built on top of the existing Spec Kit stages.
## Problem
Today, Spec Kit provides strong individual stages like:
- specify
- clarify
- plan
- checklist
- tasks
- analyze
- implement
But users who want a more autonomous "orchestrated" experience still have to decide:
- how to sequence these stages safely
- when to stop vs continue
- when to ask the user for a decision
- when issue creation should happen
- when implementation should *not* happen automatically
There are also workflow-oriented discussions in the repo around more structured orchestration and around treating clarify/checklist/analyze as more explicit parts of a standard flow.
### Proposed Solution
## Proposal
I'd like to contribute a repository example for a **Spec Kit autonomous orchestrator** that:
- starts from the user request
- runs `specify`
- runs `clarify`
- runs `plan`
- runs `checklist`
- runs `tasks`
- runs `analyze`
- stops by default after artifact generation
- only creates issues if explicitly requested
- only implements if explicitly requested
### Design goals
- artifact-first, not auto-implementation-first
- safe defaults
- explicit stop conditions
- explicit loop-back behavior when artifacts become stale
- suitable for users who want a higher-assurance, more automated workflow
## Scope
What I am proposing to contribute is one of the following, depending on maintainer preference:
1. **Documentation example**
- a documented example of a hands-off orchestrator/workflow
2. **Example workflow/agent file**
- contributed as an example/pattern in the repo
3. **Reference workflow contribution**
- if maintainers think this belongs as a first-class workflow example
## Constraints for the initial proposal
For the first version, I want to keep it intentionally conservative:
- use only `README.md` and `DESIGN.md` as in-repo guidance sources
- add a visible warning about those constraints
- no automatic issue creation unless explicitly requested
- no automatic implementation unless explicitly requested
## Non-goals
This is **not** intended to:
- force a new default behavior for Spec Kit
- replace the existing stage-by-stage workflow
- introduce hidden autonomous implementation
- assume all users want bypassed approvals
## Why I think this fits the repo
Spec Kit already has workflow support and an explicit multi-stage SDD model, and there are existing issues around higher-level orchestration and making clarification/analysis/checklist part of a more standard flow. This proposal is meant to build on that direction as an example users can adopt or adapt.
## Questions for maintainers
1. Would you prefer this as:
- docs only,
- an example workflow/agent file,
- or a more official workflow contribution?
2. Is there an existing preferred location for contributed orchestration examples?
3. If this seems in-scope, I’m happy to open a PR with the proposed example and docs.
### Alternatives Considered
_No response_
### Component
Agent integrations (command files, workflows)
### AI Agent (if applicable)
All agents
### Use Cases
## Specific use cases
- **Teams that want autonomous spec creation without autonomous coding**
A user can describe a feature once, and the orchestrator reliably runs `specify -> clarify -> plan -> checklist -> tasks -> analyze`, then stops with ready-to-review artifacts instead of immediately implementing code.
- **Repos that want a consistent artifact-first workflow for every feature request**
Instead of each contributor manually deciding which Spec Kit stages to run and in what order, the orchestrator provides a repeatable default flow with explicit gates and stop conditions.
- **Product or engineering leads preparing implementation-ready work for others**
A lead can use the orchestrator to turn a rough feature idea into a clarified spec, technical plan, and task breakdown that another engineer or agent can implement later.
- **Users who want safer automation with explicit escalation points**
The orchestrator can resolve straightforward ambiguity from repo guidance, but pauses when a real product decision is required, reducing the risk of silently making the wrong choice.
- **Issue-driven teams that want task generation before execution**
A user can request tracking-only output so the orchestrator produces tasks and optionally converts them into issues, without implementing the feature.
- **Contributors who want a reference example for advanced Spec Kit orchestration**
The repository can provide a concrete example of how to combine existing stages like `clarify`, `checklist`, and `analyze` into a higher-assurance workflow.
### Acceptance Criteria
## Acceptance Criteria
- A repo-contributed example workflow or agent is available that orchestrates these stages in order:
1. `speckit.specify`
2. `speckit.clarify`
3. `speckit.plan`
4. `speckit.checklist`
5. `speckit.tasks`
6. `speckit.analyze`
- The example has **safe default behavior**:
- stops after artifact generation unless the user explicitly requests issue creation or implementation
- does not create issues unless explicitly requested
- does not implement unless explicitly requested
- The example defines **clear loop-back behavior**:
- reruns earlier stages when `clarify`, `checklist`, or `analyze` reveal stale, incomplete, or inconsistent artifacts
- The example defines **clear escalation behavior**:
- asks the user when a true product decision is required instead of guessing
- The example is documented clearly enough that a user can understand:
- when to use it
- what it automates
- what it intentionally does not automate
- The workflow or example is reviewed and accepted by maintainers as aligned with Spec Kit’s intended stage ordering and suitable as a repository example, documentation example, or reference workflow.
### Additional Context
## Additional context
I am proposing this as an **example contribution**, not as a mandatory change to Spec Kit’s default behavior.
The main idea is to provide a repo-hosted pattern for users who want a more autonomous but still safe orchestration layer on top of existing Spec Kit stages.
### Proposed behavior at a glance
Default flow:
1. `speckit.specify`
2. `speckit.clarify`
3. `speckit.plan`
4. `speckit.checklist`
5. `speckit.tasks`
6. `speckit.analyze`
7. stop and report readiness
Optional follow-up stages:
- `speckit.taskstoissues` only if the user explicitly requests issue creation
- `speckit.implement` only if the user explicitly requests implementation
### Why this seems useful
Spec Kit already provides the building blocks for a strong staged workflow, but users still have to decide how to safely compose them. A contributed example could help demonstrate:
- an artifact-first workflow
- explicit stop conditions
- explicit loop-back behavior
- safer defaults around issue creation and implementation
### Constraints I would keep in the first version
- use only `README.md` and `DESIGN.md` as in-repo guidance sources
- add a visible warning about those constraints
- avoid relying on other repo documents unless explicitly provided by the user
- keep issue creation and implementation opt-in
### Example user scenarios
- “Take this rough feature request and produce a reviewed-ready spec, plan, tasks, and analysis, but do not implement anything.”
- “Generate the full set of artifacts, then convert tasks into issues for tracking.”
- “Run the artifact flow first, and only implement if I explicitly confirm.”
### What I am asking maintainers
I mainly want guidance on whether this is best contributed as:
- a docs example
- an example workflow/agent file
- or a more official reference workflow
If this is considered in-scope, I’d be happy to open a PR with the proposed example and documentation.
贡献指南
调研方向
先阅读 README.md 和 DESIGN.md,然后检查现有 Spec Kit 阶段如何为 agent 工作流提供支持。在实现 specify-through-analyze 序列之前,与维护者确定示例的形式和位置,包括选择加入的 issue 创建与实现、回环和升级。已完成的标准是:已接受的示例及其使用指南满足所有列出的验收标准。
由索引模型根据 Issue 内容生成。
评估
- 领域
- ai, tooling
- Issue 类型
- 功能
- 难度
- 5/5
- 预计耗时
- 一周以上
- 活跃度
- 活跃
- 描述清晰度
- 基本清楚
- 新手友好度
- 35/100