About planning-and-task-breakdown
planning-and-task-breakdown is a stack-agnostic skill for turning specifications into repository plans and, when useful, tracked tasks. It requires read-only exploration and planning before implementation.
Key features
- 01Ground plans in workflow/SPEC.md
- 02Explore dependencies in read-only mode
- 03Write workflow/PLAN.md with verification
- 04Create workflow/TASKS.md when warranted
- 05Check coverage, dependencies, and scope
Use cases
- 01Breaking down a non-trivial change
- 02Planning work across deliverables
- 03Preparing independent tasks for parallel work
- 04Defining verification before implementation
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ alexbruno/ dev-skills --skill planning-and-task-breakdownAfter installing, type/planning-and-task-breakdownin Claude Code to confirm it loaded.
Details
From the project READMEPlanning & Task Breakdown
Decompose specifications into clear execution steps (workflow/PLAN.md) and, when warranted, small verifiable tasks (workflow/TASKS.md) before writing any implementation code.
Language & Architecture Agnostic: This skill applies to any technology stack, domain, or project structure. Reason in terms of deliverables, boundaries, dependencies, and verification—never assume specific languages, frameworks, or architectural patterns.
Core Rule
Plan in read-only mode. Explore the repository, map dependencies, and write the plan before modifying any deliverable files. Do not start implementing while planning.
When to Apply
- Translating
workflow/SPEC.mdinto an actionable execution plan (workflow/PLAN.md). - Work feels too large, ambiguous, or multi-faceted to execute in a single pass.
- Multiple independent deliverables or parallel/asynchronous agent executions are possible.
- Skip formal breakdown for trivial, single-step edits where planning adds no value.
Planning Process
1. Ground in workflow/SPEC.md
- Read
workflow/SPEC.md(orworkflow/<change-slug>/SPEC.mdfor scoped initiatives).- If
SPEC.mddoes not exist yet for a non-trivial change, create and align onSPEC.mdfirst.
- If
- Confirm goals, non-goals, constraints, and acceptance criteria are unambiguous. Surface open questions before committing to a plan.
2. Explore & Map Dependencies (Read-Only)
- Inspect the relevant areas of the repository (preferring indexed context if
codebase-indexingis active). - Identify affected components, shared boundaries, and external constraints.
- Map the dependency graph: determine what blocks what, which steps must run sequentially, and which units are truly independent.
3. Author workflow/PLAN.md
Create or update workflow/PLAN.md (initializing workflow/ if missing):
- Reference
SPEC.mdrequirements rather than duplicating them. - Order steps logically—tackle high-risk or high-uncertainty unknowns early.
- Define a concrete Verification Strategy (automated checks, validations, or manual verification) for each step and for the overall change.
- Note risks and rollback considerations.
4. Shard into workflow/TASKS.md (Conditional)
Create workflow/TASKS.md only when at least one spec-driven-workflow trigger applies:
PLAN.mdis too large or comprehensive to execute cleanly in a single pass.PLAN.mdspans two or more distinct scopes, task types (feature,bugfix,refactor,docs), or independent deliverables that benefit from separate tracking.- Work can be split into independent units executed in parallel or asynchronously by different agents or contributors.
When creating TASKS.md, ensure every task has:
- Unique ID & Type (e.g.,
T1,feature|bugfix|refactor|docs|chore). - Explicit Dependencies: Clearly state blocking task IDs (or
None). - Disjoint Scope: Assign non-overlapping files or modules to parallel tasks to prevent merge conflicts.
- Objective Done Criteria: Concrete, verifiable conditions that prove the task is complete.
Quality Checklist
Before handing off PLAN.md (and TASKS.md) for execution:
- Every requirement and acceptance criterion in
workflow/SPEC.mdis covered by at least one step or task. - No circular dependencies exist, and parallel tasks have non-overlapping scopes.
- Every step/task is small enough to implement and verify in a single focused pass.
- No implementation code has been written prematurely during the planning phase.