Skip to content

Latest commit

History

History
74 lines (50 loc) 路 3.35 KB

File metadata and controls

74 lines (50 loc) 路 3.35 KB

Hands-Off Orchestration Example

This example shows how to ask an AI coding agent to run a hands-off, artifact-first Spec Kit flow without moving into issue creation or implementation by default.

Use this pattern when you want a repeatable planning pass that produces reviewed artifacts first:

/speckit.specify -> /speckit.clarify -> /speckit.checklist -> /speckit.plan -> /speckit.tasks -> /speckit.analyze

This ordering matches the Quick Start Guide recommended workflow, where /speckit.checklist is a requirements-quality gate before /speckit.plan.

The key rule is that the agent stops after analysis unless the user explicitly asks it to continue.

Note

This conservative example uses only README.md and DESIGN.md as in-repo guidance sources unless the user explicitly provides more context.

Guardrails

  • Generate and review artifacts before creating issues or writing code.
  • Ask the user when a product decision is required.
  • Loop back to an earlier stage when a later stage finds stale, incomplete, or inconsistent artifacts.
  • Use only README.md and DESIGN.md as in-repo guidance sources unless the user explicitly provides more context.
  • Do not run /speckit.taskstoissues unless the user explicitly requests issue creation.
  • Do not run /speckit.implement unless the user explicitly requests implementation.

Example Agent Prompt

Copy this prompt into your agent after initializing a Spec Kit project:

Run an artifact-first Spec Kit orchestration flow for the following feature:

<describe the feature here>

Follow this sequence:

1. Run /speckit.specify for the requested feature.
2. Run /speckit.clarify to identify ambiguity. If a real product decision is required, stop and ask me before continuing.
3. Run /speckit.checklist and fix requirement-quality gaps by looping back to /speckit.clarify or /speckit.specify as needed.
4. Run /speckit.plan after the specification passes the requirements checklist.
5. Run /speckit.tasks after the plan and checklist are consistent.
6. Run /speckit.analyze to check cross-artifact consistency.
7. Stop after analysis and report the generated artifacts, open questions, and recommended next step.

Use only README.md and DESIGN.md as in-repo guidance sources unless I explicitly provide more context.
Do not create GitHub issues unless I explicitly ask for issue creation.
Do not implement tasks unless I explicitly ask for implementation.

Optional Follow-Up Prompts

After reviewing the generated artifacts, continue with one of these explicit prompts:

Run /speckit.taskstoissues to create GitHub issues from the generated tasks, but do not implement them.
Run /speckit.implement to implement the generated tasks now.

Keeping issue creation and implementation as separate follow-up prompts makes the default flow safe for planning, design review, and handoff scenarios.

When To Use This Pattern

Use this pattern when:

  • A team lead wants implementation-ready artifacts before assigning work.
  • A project wants consistent specs, plans, and tasks for every feature request.
  • The feature has enough ambiguity that clarification and checklist passes are useful.
  • You want automation to prepare work, but not silently create issues or write code.

For smaller experiments, the shorter workflow in the Quick Start Guide may be enough.