Work Breakdown Structure

Overview

Read a plan document, analyze the scope, propose a phase structure for approval, then produce phase documents and a phases.md execution tracker. The original plan document is left untouched — it is a stable reference.

Each phase must be independently committable — it should leave the codebase in a working state. Phases flow forward: earlier phases unblock later ones, never the reverse.

Step 1 — Locate the plan

If the user provided a path argument, use it. Otherwise:

• Search for plan files matching plans/**/plan.md
• If exactly one exists, use it
• If multiple exist, list them and ask the user which one to break down. If operating autonomously, choose the only incomplete plan; if all are incomplete, choose the most recently modified one and note the assumption.
• If none exist, tell the user to run /plan-init-v1 <task> first, or provide any markdown file containing a goal and scope description

Read the plan document in full.

If the plan carries a Format: v2 marker, stop: v2 plans are broken down by /plan-phase (which writes an execution.md tracker), not this v1 skill. Tell the user to run /plan-phase instead. (This mirrors the v2 skills, which refuse a v1 plan — the guard runs both ways.)

Handling non-plan-init-v1 input

If the file was not produced by /plan-init-v1, look for equivalent content:

If a section is missing entirely, note what couldn't be found and either:

• Ask the user to clarify (if the gap is material — e.g. no goal at all)
• Make a reasonable inference and proceed, noting the assumption

Do not refuse to proceed just because the format differs from plan-init-v1 output.

Output directory: phase documents and phases.md are always written into the same directory as the source plan file — the plans/<slug>/ paths used in Steps 5–7 refer to that directory. When the plan lives outside plans/ (e.g. docs/proposal.md), write the phase docs and phases.md alongside it (docs/), never into an invented plans/<slug>/, so /plan-run-v1 finds them next to the plan.

Step 2 — Explore for breakdown context

Based on the plan's affected areas (or equivalent scope section), explore the codebase:

• Read the key files that will change
• Identify natural seams: what can be done independently, what has dependencies
• Look at existing test files to understand what tests currently exist vs. need to be written
• Note any migration concerns (DB schema, API contracts, translations, etc.)

Verify every assumption against the actual codebase before designing phases. Do not assume a file exists or a path is correct — check it. If the plan's Affected Areas list contains paths that don't exist or have moved, note the discrepancy.

Use this exploration to inform phase granularity. Phases should be:

• Small enough to commit independently (1–6 hours of focused work)
• Large enough to be meaningful (not a single line change unless it's a critical gate)
• Ordered so each builds on the last without breaking anything

Step 3 — Design the phases (internal)

Think through the full sequence before presenting anything to the user. Consider:

1. Foundation first — schema changes, new models, interface definitions before their consumers
2. Tests are mandatory, not optional — every phase that introduces new behavior must include tests for that behavior. For API endpoints, business logic, and utilities: prefer TDD (failing test written before implementation). For UI and wiring code: tests written alongside. No phase is complete without its tests passing.
3. Backend before frontend (usually) — or at least the API contract before the UI
4. Risky changes isolated — put anything with blast radius in its own phase
5. Final phase = verification gate — last phase runs all success criteria from the plan

Target 3–8 phases. Too few means each phase is too risky; too many means overhead.

For each phase determine:

• A short human-readable name (title case, e.g. "Add Schema Migration")
• A kebab-case filename slug (e.g. add-schema-migration)
• A one-sentence goal
• Entry criteria: what must be true before this phase can start
• A rough list of tasks (3–8 bullet points — specific but not yet fully detailed)
• Exit criteria: the formal definition of done for this phase (separate from the task checklist)
• The verification command(s) to run at the end

Step 4 — Present proposed phases and ask for approval

Before writing any files, show the user the proposed breakdown in a single message:

Here's the proposed breakdown for "<plan title>" — N phases:

Phase 1: <Name>
  Goal: <one sentence>
  Entry: <what must be true before this phase starts>
  Tasks: <3–5 bullet points of what this phase covers>
  Verify: <test command>
  Exit: <formal done criteria>

Phase 2: <Name>
  ...

Does this breakdown look right? A few things you can tell me:
- Add, remove, or merge phases
- Move tasks between phases
- Change the scope of any phase
- Adjust the order

Reply "looks good" to proceed, or describe any changes.

Wait for the user's response. If they request changes, revise the breakdown and re-present it. Repeat until they confirm. Do not create any files until the user approves the structure.

If operating autonomously (no user available), proceed with the internally designed phase breakdown and note the assumption that it was not user-reviewed.

Step 5 — Write phase documents

Check for an existing phases.md first — before writing anything. A tracker with phases already checked off (- [x]) records execution progress that a fresh breakdown would destroy, and the phase documents below are written before phases.md is, so a guard that waits until Step 6 has already let Step 5 overwrite the phase documents whose progress that tracker is pointing at. The half-executed plan is the common case, and it is exactly the one that check has to survive.

So: if phases.md already exists in the plan's directory, or the directory holds any phase-*.md at all, stop and ask the user whether to regenerate the breakdown (losing what is there) or keep it. Write freely only when neither is present — not merely when none of the tracker's phases is ticked.

The phase-*.md-without-phases.md case is the one a tracker-only guard waves straight through, and the one where the damage is worst: an earlier run of this skill stopped between Step 5 and Step 6, so there is no tracker to resume from and nothing looks wrong until the approved phase documents have already been overwritten. Do not assume an unfinished directory is a scratch one. An unticked tracker beside phase documents whose task boxes are filling in is precisely what a run interrupted inside its first phase looks like, and "no completed phases" would wave that case through as if it were a clean slate. Progress lives in the phase documents too; the tracker is a derived index and is the last thing a run updates.

For each approved phase N, create plans/<slug>/phase-<NN>-<name>.md (zero-padded number, e.g. phase-01-add-schema-migration.md).

Each phase document uses this structure:

# Phase <N>: <Human-Readable Name>

_Status: pending_

## Goal

<One sentence: what this phase accomplishes and why it matters.>

## Entry Criteria

Before starting this phase, confirm:
- [ ] <Prior phase committed, pushed to origin, and verified, or "This is the first phase">
- [ ] <Any specific precondition — e.g. "migration X applied", "API shape agreed">

## Tasks

- [ ] <Specific task: what file, what change, what outcome>
- [ ] <Another specific task>
- [ ] <...>

## Tests

_For logic, API endpoints, and utilities: write failing tests before implementation (TDD).
For UI and wiring: write tests alongside the code._

- [ ] `<path/to/test_file>` — <what behavior it covers>
- [ ] <additional test file if needed>

## Verification

Run these after completing all tasks:

```bash
<test command 1>
<test command 2 if needed>
```

Also verify manually:
- <Specific thing to check that tests won't catch>

## Exit Criteria

This phase is complete only when ALL of the following are true:
- [ ] Every task above is checked off
- [ ] All tests listed in the Tests section are written and passing
- [ ] No previously passing tests have regressed
- [ ] All verification commands pass with no failures
- [ ] Run the `cyw` skill (or equivalent manual review) — finds zero issues
- [ ] <Any additional condition specific to this phase>
- [ ] phases.md phase checkbox updated to `[x]`

## Commit

```
<suggested commit message — imperative, under 72 chars>
```

Use real file paths from the codebase. Tasks should be specific enough that another engineer could follow them without re-reading the plan. Follow all conventions defined in the project's CLAUDE.md / AGENTS.md (whichever exists).

Step 6 — Write phases.md

The overwrite check happened in Step 5, before any phase document was written.

Create plans/<slug>/phases.md — the execution tracker. Do not modify the plan. Its header link names the plan file you actually read, which is beside it; plan.md is only the usual name, not the guaranteed one.

# Phases: <human-readable title>

_Execution tracker for [`<plan filename>`](./<plan filename>)_

## Status

| Field | Value |
|---|---|
| Phase | Phase 1 of N — <Phase 1 Name> |
| State | Ready to execute |
| Blocker | None |
| Last updated | <date> |

## Phases

- [ ] [Phase 1: <Name>](./phase-01-<name>.md)
- [ ] [Phase 2: <Name>](./phase-02-<name>.md)
- [ ] ...

Step 7 — Report to the user

Print a summary:

• Location of phases.md and number of phase files created
• One-line description of each phase
• Total checklist items across all phases
• Next step: "Run /plan-run-v1 <the plan file you read> to execute all phases." — its real path, not a literal plans/<slug>/plan.md, which names nothing when the plan came from somewhere else.