plan-tasks

name	plan-tasks
description	Break a PRD or feature into numbered implementation tasks
user-invocable	true
argument-hint	<prd-or-feature>

Task Breakdown

Break a PRD or feature description into concrete, ordered implementation tasks.

Process

1. Read the source material:

   • If a PRD path is given, read it
   • If a feature name is given, look for a matching PRD in docs/prds/
   • Read docs/architecture.md and CLAUDE.md for project conventions

2. Identify work items by decomposing the feature into atomic, implementable tasks.

3. Output a numbered task list with each task containing:

   Field	Description
   #	Sequential number
   Title	Short imperative description (e.g., "Add git_status Tauri command")
   Description	What to implement, acceptance criteria
   Complexity	S (< 30 min), M (30 min - 2 hrs), L (2+ hrs)
   Category	backend (Rust/Tauri), frontend (React/TS), or both
   Dependencies	Task numbers that must complete first (e.g., "Depends on #1, #2")
   Files	Key files to create or modify

4. Order tasks so dependencies come first. Group by layer when possible:

   • Backend (Tauri commands, Rust types) first
   • State management (Zustand stores) second
   • UI components last

5. Include a header table immediately after the # ... title:

   |  |  |
   | --- | --- |
   | **Date** | YYYY-MM-DD |
   | **Status** | Not started |
   | **PRD** | [slug](../prds/YYYY-MM-DD-slug.md) |
   | **Total** | 12 tasks: 4S, 5M, 3L |
   | **Suggested order** | Backend (#1-#4) → State (#5-#6) → UI (#7-#12) |
   

   Follow the table with any risks or open questions.

6. Save the task list to docs/tasks/YYYY-MM-DD-<slug>-tasks.md where:

   • Date is today's date
   • Slug matches the PRD slug (e.g., PRD docs/prds/2026-03-11-model-metadata-enrichment.md → tasks file docs/tasks/2026-03-11-model-metadata-enrichment-tasks.md)
   • If no PRD exists, derive the slug from the feature name in lowercase-kebab-case

7. Update the PRD's header table with a two-way link back to this new tasks file:

   • Add or update a **Tasks** row: | **Tasks** | [slug-tasks](../tasks/YYYY-MM-DD-slug-tasks.md) |
   • This establishes the bidirectional PRD ↔ tasks navigation. Readers should be able to jump in either direction without searching.

8. Update the source research doc if this task breakdown traces back to a research file in docs/research/:

   • Find the research file (check the PRD's header for a research reference, or match by topic)

   • Add or update a Tasks row in the research doc's pipeline table:

     | Stage | Link | Status |
     | --- | --- | --- |
     | PRD | [slug](../prds/YYYY-MM-DD-slug.md) | Complete |
     | Tasks | [slug-tasks](../tasks/YYYY-MM-DD-slug-tasks.md) | Not started |
     

   • Valid statuses: Not started, In Progress, Complete, Partial, Abandoned, Not planned

   • If no research doc exists for this feature, skip this step

9. Present a brief summary to the user with the file path and task count.

10. Log observations to .claude/skill-feedback.md if the decomposition showed issues (duplicates, wrong sizing, missing dependencies, order problems, unclear acceptance criteria). Format per /retrospect-skills. Both user and agent contribute.

Guidelines

• Tasks should be small enough to complete in one sitting
• Each task should be independently testable
• Include "write tests" as explicit tasks where applicable
• Flag tasks that touch shared infrastructure or have high blast radius
• Reference existing patterns — don't reinvent (e.g., "Follow the pattern in commands/file.rs")