# Spec Loop Plan Work Breakdown > Create or revise file-based Spec Loop work breakdown artifacts for task-file work, including splitting a requested single item into subtasks or multiple task files / backlog items when planning or design shows that split is needed. - Skill: `dpolivaev/spec-loop-plan-work-breakdown` (Agent Skill) - Install (CLI): `npx skillmds@latest add dpolivaev/spec-loop-plan-work-breakdown` - Raw SKILL.md: https://api.skillmd.com/api/skills/dpolivaev/spec-loop-plan-work-breakdown/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Product & Planning - Author: dpolivaev (https://skillmd.com/u/dpolivaev) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/dpolivaev/spec-loop-plan-work-breakdown --- Use this skill for file-based Spec Loop work breakdown on the task-file path. It is a planning companion to `spec-loop-plan-task`; it does not replace [planning-form selection](../spec-loop-plan-task/planning-form-selection-guidance.md), clarification, task-file rules, or approval gates. Use it after planning-form selection chooses `task file with subtasks` or `multiple task files / backlog items`. Also use it when revising an existing file-based work breakdown, or when later planning or design shows that a requested single task or subtask should be split. If this skill is loaded before planning-form selection is complete, do not draft breakdown artifacts. If first classification is incomplete, or if later planning or design may change the viable forms, return to [spec-loop-plan-task](../spec-loop-plan-task/SKILL.md). Otherwise follow [guidance](../spec-loop-plan-task/planning-form-selection-guidance.md) first. If this need appears during chat-only/fileless planning, switch or promote to the task-file path before creating the breakdown. Chat-only content is not durable project state outside the current chat, so work breakdown state must live in a task file. In the sections below, `(sub)task` means a task or subtask in the work breakdown. ## Scope and motivation gate Do not draft or revise a work breakdown until the whole work being split has clear Scope and Motivation. For initial work breakdown and material revisions to an existing work breakdown, apply the clarification skill's mandatory decision screening to: - the whole work's Scope and Motivation; - each proposed (sub)task Scope and Motivation; and - any Constraint included in a (sub)task or needed to decide (sub)task boundaries, ordering, releasability, or coherence. Do not use work breakdown to discover, clarify, or decide Constraints that only affect later Design or Test specification inside an already coherent (sub)task. If an unknown constraint affects (sub)task boundaries, ordering, releasability, or coherence, treat it as a breakdown blocker; otherwise defer it until that (sub)task becomes current. If Scope or Motivation is unclear, do only enough research to frame the gap, then use [spec-loop-clarify-task/SKILL.md](../spec-loop-clarify-task/SKILL.md) before drafting breakdown items. Do not create placeholder implementation (sub)tasks for unclear future work. For work breakdown, clarify only enough to make (sub)task boundaries, ordering, and coherence sound. Defer detailed rule, API, design, and test clarification until the (sub)task is current. ## Standalone (sub)task rule Every (sub)task in a work breakdown must stand on its own by default. For implementation work, that means the (sub)task must be releasable. A releasable implementation (sub)task can be implemented, verified, reviewed, and accepted on its own while leaving the project coherent if no later sibling is implemented. It must include its tests and any supporting config, migration, documentation, glossary, or operational updates needed to stand on its own. This is a validity criterion, not a recommendation to release each (sub)task independently. For feature implementation, choose (sub)task boundaries by slicing the overall work into releasable vertical slices when a coherent split exists. Use separate task files/backlog items when separate release decisions are advisable. Use subtasks when the resulting subtasks can safely stand alone but separate release decisions are not advisable. Do not create non-releasable implementation (sub)tasks unless the User explicitly opts out. If the User opts out, record the opt-out in the affected (sub)task's Scope. Forbidden by default: - scaffolding-only, setup-only, model-only, logic-only, API-only, persistence-only, UI-only, test-only, or docs-only implementation (sub)tasks that do not stand alone; - a preparation-heavy implementation (sub)task with no observable accepted-result difference or other immediately demonstrable coherent capability when a smaller behavior-first vertical slice is viable; - a "foundation" subtask that only enables later behavior but has no independently reviewable result; and - splitting implementation and tests for the same behavior into separate subtasks. A standalone refactoring makes no intended externally observable behavior change and is still acceptable if merged even if the later feature item is never implemented. Allowed when they stand alone: - infrastructure or setup work that leaves a coherent build or runtime capability; - documentation, test-hardening, migration, or cleanup work whose standalone artifact is the accepted deliverable; - standalone refactoring; and - a layer-focused change that is itself a complete reviewable behavior or operational capability, not merely a dependency for a later sibling. ## Fast-payoff preference Prefer the earliest coherent implementation (sub)task that shows visible executable behavior or another immediately demonstrable coherent capability. For feature work, that usually means the first releasable vertical slice. Defer setup, generation, storage, orchestration, and generalization until they affect the accepted result of that (sub)task, unless they are required for coherence or independent acceptability. If the same behavior can be reviewed first with deterministic setup, prefer that slice. If a candidate (sub)task still carries substantial plumbing or generalization with too little visible result, or internal choices it cannot yet expose, split it again unless that would break independent acceptability. Do not pull speculative flexibility or future-proofing into an earlier (sub)task when a narrower one can already stand alone. ## Work breakdown detail level For an initial (sub)task created from scratch, include only the title, required identifier/status metadata, and these planning sections: - Scope; - Motivation; and - Constraints, only for already known limits that materially affect the (sub)task. For subtasks, keep the normal `- **Status:**` line because it is lifecycle metadata, not planning content. For research, spike, prototype, catalog, or proof (sub)tasks, Scope states the investigation boundary and expected output; Motivation states what it blocks or enables. Do not newly draft Briefing, Research, Scenario, Glossary, Analysis, Design, or Test specification during work breakdown. Complete those planning sections only after the (sub)task becomes current under `spec-loop-plan-task`. Do not create Findings or Implementation notes during work breakdown. Findings is filled only after approved investigation work produces reviewed output. Implementation notes follow `spec-loop-implementation-flow`. Include a Constraints section only for already known limits that materially affect the (sub)task. Do not create placeholder Constraints. The exception is restructuring existing planned work, including converting a no-subtask task into subtask form. Existing section content may be moved to the correct resulting (sub)task, but new content for those sections must not be drafted during breakdown. Before execution approval on the task-file path, complete the sections required for the current task under `spec-loop-plan-task`. If the task uses subtasks, apply this to the current subtask plus needed task-level context. ## Blocking unknowns If a blocking unknown appears during work breakdown, use [spec-loop-clarify-task/SKILL.md](../spec-loop-clarify-task/SKILL.md) when existing evidence or User engagement can resolve it. If the blocker still needs a durable decision, route ADR work. If it needs new evidence, create a research, spike, prototype, catalog, or proof (sub)task only when that investigation requires explicit planning, separate tracking, or a reviewable output. If accepted investigation findings show that a durable architecture or policy decision is needed, route ADR work before planning dependent implementation (sub)tasks. Do not create downstream implementation (sub)tasks or placeholders that depend on the unknown result. If blocking work is needed, pause work breakdown. After that work is accepted, continue the work breakdown from the new evidence or decision. If converting a no-subtask task to subtask form, adding a new subtask after earlier subtasks already contain full section content, or making the task file too large to use safely, use [spec-loop-compact-task-file/SKILL.md](../spec-loop-compact-task-file/SKILL.md) and then resume this skill. ## Work breakdown construction When drafting a work breakdown: - name each (sub)task by its delivered outcome, not by a technical layer; - when creating backlog task files, number them with readable three-digit prefixes local to the containing folder; - prefer the smallest sequence where each (sub)task remains independently acceptable, each implementation (sub)task remains releasable, and earlier (sub)tasks follow the Fast-payoff preference above; - create research, spike, prototype, catalog, or proof (sub)tasks only when the investigation itself requires explicit planning, separate tracking, or a reviewable output; - for implementation (sub)tasks, include all cross-layer work needed in that same (sub)task; - if the refactoring is standalone and the change remains one overall task, put it in its own subtask; - if the refactoring is standalone and separate release decisions are advisable, use a separate task file or backlog item; - if the refactoring is not standalone, keep it inside the feature (sub)task; - keep future (sub)tasks created from scratch to title, required identifier/status metadata, Scope, Motivation, and any already known relevant Constraints until they are current, but make the standalone result clear; - place supporting docs, glossary changes, config, migration, and tests in the same (sub)task as the behavior they support unless they are standalone deliverables; and - if a candidate implementation (sub)task is not releasable, reshape it, merge it with an adjacent (sub)task, or ask the User whether to opt out explicitly. ## Final self-check Before presenting or saving a file-based work breakdown, verify: - each implementation (sub)task can reach `review` on its own; - each implementation (sub)task has or will receive its own automated tests; - for implementation (sub)tasks without an explicit non-releasable opt-out, after each completed (sub)task the resulting software state would be coherent and releasable if no later sibling were implemented; - earlier implementation (sub)tasks, especially the first, satisfy the Fast-payoff preference above and do not pull setup, generation, storage, orchestration, or generalization forward without need; - no (sub)task is named only for a layer or preparation activity unless it is independently acceptable; - any non-releasable implementation (sub)task has explicit User opt-out recorded; - each research, spike, prototype, catalog, or proof (sub)task shows why separate tracking is needed; - no downstream implementation (sub)task depends on an unresolved unknown; - new backlog task files created by this breakdown have readable three-digit prefixes local to the containing folder; - each (sub)task created from scratch contains only the title, required identifier/status metadata, Scope, Motivation, and already known relevant Constraints when present; - each refactoring-only (sub)task is a standalone refactoring; - separate task files or backlog items are used for standalone refactoring only when separate release decisions are advisable; - fuller section content appears only in an unchanged pre-existing current task, in an unchanged pre-existing current subtask, or because it was moved from an existing (sub)task during restructuring, including conversion from no-subtask task to subtask form.