specify-prepare

name: specify-prepare description: Orchestrates the pre-implementation pipeline (specify -> plan -> tasks). Supports single features, multi-feature dispatch, and auto-discovery from docs. Uses parallel sub-agents (via git worktrees) for efficiency.

Specify Prepare Pipeline (Parallel Pre-Implementation)

This skill automates the "pre-implementation" phase for the specify workflow in Gemini CLI. It translates broad feature descriptions or existing documentation into concrete technical specifications (spec.md), implementation plans (plan.md), and actionable tasks (tasks.md).

It efficiently handles single features, explicit multi-feature dispatch, and auto-discovery of pending issues by utilizing git worktrees and Gemini sub-agents to process features in parallel.

1. Operating Modes

First, determine the operating mode based on the user's input:

A. Auto-Discover (--auto-discover)

1. Scan Issues: Use the glob tool to find all markdown files in docs/issues/**/*.md.
2. Filter Pending: Read the files. Identify issues that have a status like pending, todo, planned, or 미구현 (or missing status). Ignore those marked as done, in-progress, etc.
3. Confirm with User: Use the ask_user tool (with type: choice and multiSelect: true) to list the discovered candidates and ask the user which ones they want to prepare.
4. Extract Descriptions: For each selected issue, extract a concise feature description (ID + Title + core scope).
5. Proceed to Dispatch: Treat the selected descriptions as a Multi-Feature Dispatch.

B. Multi-Feature Dispatch (Multiple descriptions provided)

If the user provides multiple feature descriptions separated by commas or quotes, proceed directly to Section 2: Parallel Dispatch.

C. Single Feature

If the user provides a single feature description, proceed directly to Section 2: Parallel Dispatch, but with only one item.

2. Parallel Dispatch

For each feature description to be processed, perform the following steps:

1. Create Branch & Directory: Run the following command to initialize the feature branch and get the paths:

   .specify/scripts/bash/create-new-feature.sh --json "<feature_description>"
   

   Parse the JSON output to get BRANCH_NAME, SPEC_FILE, and FEATURE_DIR.

2. Setup Git Worktree: Create a temporary git worktree to isolate the preparation process and avoid file locking conflicts:

   # Use the branch name for the worktree directory name to keep it organized
   git worktree add ../<project-name>-prep-<BRANCH_NAME> <BRANCH_NAME>
   

3. Launch Sub-Agent (Worker): Use the run_shell_command tool with is_background: true to run a Gemini sub-agent inside the worktree. Ensure it automatically bypasses interactive checkpoints (like the old --auto mode).

   Example command:

   cd ../<project-name>-prep-<BRANCH_NAME> && gemini "Activate the specify-prepare skill and run the Worker Pipeline for feature: '<feature_description>'. Target directory is '<FEATURE_DIR>'. Write spec.md, plan.md, and tasks.md based on .specify/templates/. Commit changes when done."
   

   Keep track of the background PIDs.

4. Monitor & Cleanup: Wait for all sub-agents to complete their work. Then, for each worktree:

   git worktree remove ../<project-name>-prep-<BRANCH_NAME>
   

   (The <BRANCH_NAME> remains in the main repository, ready for implementation).

5. Final Report: Output a summary table to the user showing: Feature Description, Branch Name, and Readiness Status.

3. Worker Pipeline (For Sub-Agents)

If you are a sub-agent executing inside a worktree (instructed by Step 2.3), follow these steps strictly (no user confirmation required):

1. Context Gathering:

   • Load .specify/memory/constitution.md (if exists).
   • Read docs/**/*.md to understand domain entities, the tech stack, and constraints.

2. Generate Specification (spec.md):

   • Read .specify/templates/spec-template.md.
   • Create <FEATURE_DIR>/spec.md using the gathered context and feature description. Resolve any ambiguities internally using best practices.

3. Generate Plan (plan.md):

   • Run .specify/scripts/bash/setup-plan.sh --json.
   • Read .specify/templates/plan-template.md.
   • Create <FEATURE_DIR>/plan.md addressing architectural design, data models, and API contracts. Ensure your plan aligns with the constitution principles.

4. Generate Tasks (tasks.md):

   • Read .specify/templates/tasks-template.md.
   • Create <FEATURE_DIR>/tasks.md breaking down the plan into a checklist format (- [ ] [TaskID] ...). Group into standard phases: Phase 1 (Setup), Phase 2 (Foundational), Phase 3 (UI/Logic), Phase 4 (Polish). Identify tasks that can be parallelized with [P].

5. Update Agent Context:

   • Run the script to update context files for Gemini:

   .specify/scripts/bash/update-agent-context.sh gemini
   

6. Update Documentation:

   • Call the activate_skill tool with the name specify-update-docs.
   • Following its instructions, update the corresponding feature file in docs/issues/**/*.md to change its status to prepared and add the specs/ file links.

7. Commit Changes: Stage and commit the generated specification files and the updated docs:

   git add <FEATURE_DIR> docs/issues/
   git commit -m "chore: Prepare specifications and update docs for <feature_description>"
   

8. Update Issue Status:

   • If this feature was generated from an existing issue in docs/issues/, invoke the specify-docs skill to update the status of that issue to in-progress.
   • Stage and commit this documentation update to the feature branch.