speckit-workflow（规格套件工作流程）

Speckit Workflow Skill 🦞

This is the master orchestrator for the SpecKit for OpenClaw engineering workflow.

⚠️ Security & Credentials

This workflow is designed to automate git operations to save progress and isolate work.

Agent Requirement: Before starting the workflow or initializing a new project, you MUST ask the user:

"Do you want to enable automated git commit, git push, and branch creation for this project? (Yes/No)"

• If YES: Proceed with automated commits, pushes, and branch creation as described in the workflow. Ensure you have write access.
• If NO: You MUST NOT perform any git operations (commits, pushes, or branch creation). You will only write files locally. The user is responsible for version control.

🚀 Step 1: Initialization

If this project has not been initialized with SpecKit yet, you MUST ensure the .specify/ directory is present in the project root.

Initialization Instructions for Agent:

1. Ask for Git Permission (see above).
2. Check if .specify/ exists in the project root.
3. If missing, copy the .specify/ directory from this skill package (located at ./.specify/) to .specify/ in the project root.

- Create the destination directory if it doesn't exist.

1. Confirm successful initialization.

🔄 Resuming Workflow

Before starting or when returning to a project, you MUST determine the current state by checking for the existence of SpecKit artifacts:

1. Check for Initialization: Verify if .specify/ exists.
2. Determine Current Phase:

- If .specify/memory/constitution.md exists -> Constitution complete. - If specs/<feature>/spec.md exists -> Specify complete. - If specs/<feature>/plan.md exists -> Plan complete. - If specs/<feature>/tasks.md exists -> Tasks complete. - If tasks in tasks.md are partially marked [X] -> Implementation in progress.

Always resume from the first incomplete phase in the Execution Order.

🏗️ Step 2: Workflow Orchestration

Once initialized, you MUST follow the canonical Spec-kit order by delegating to sub-agents.

When to Create a Sub-Agent:

For each phase below, you should create or delegate to a specialized sub-agent that uses the corresponding skill found in the subskills/ directory relative to this skill file (./subskills/).

The Execution Order:

1. 📜 Constitution: Delegate to speckit-constitution. Establish code quality, testing standards, and architectural constraints.
2. 📝 Specify: Delegate to speckit-specify. Transform requirements into a formal spec.md.
3. 🔍 Clarify (Optional): Delegate to speckit-clarify if the specification is ambiguous.
4. 🏗️ Plan: Delegate to speckit-plan. Derive technical design and implementation architecture.
5. 📋 Tasks: Delegate to speckit-tasks. Break the plan into actionable task lists.
6. ⚖️ Analyze (Optional): Delegate to speckit-analyze for cross-artifact consistency.
7. 👷 Implement: Delegate to speckit-implement.

👷 Implementation Session Management

When executing the speckit-implement phase, you MUST:

• Isolate Context: Trigger a new agent session for implementation to ensure focus.
• Dynamic Task Chunking: Group tasks from tasks.md dynamically based on requirements and complexity.

- If tasks are small/simple, group 3-5 tasks (e.g., T001 to T005). - If tasks are complex, group 1-2 tasks.

• Sub-Agent Execution: For each chunk, delegate to a sub-agent using speckit-implement.
• Commit & Push: After each successfully completed chunk, the sub-agent MUST commit and push the changes to the repository.
• Mark Completion: Ensure the sub-agent marks tasks as complete [X] in tasks.md before returning.
• Avoid Over-grouping: Do not group too many tasks in a single sub-agent session to maintain precision and manageable diffs.

User Provided Context

{{user_provided_context}}

Use this context to start or resume the workflow. If the project is not yet initialized, begin with Step 1: Initialization.