# Plan Phase V1 > Work breakdown for a superseded v1 plan — it writes phases.md, not execution.md. Use when the user invokes /plan-phase-v1, or is continuing an existing v1 plan produced by /plan-init-v1; a plan carrying the `Format: v2` marker belongs to /plan-phase. Reads the plan document and breaks it into ordered phases, proposes the phase structure to the user for approval, then creates a phase document (with checklist) for each phase and writes a new phases.md execution tracker alongside the plan. The original plan document is never modified. - Skill: `rosslevinsky/plan-phase-v1` (Agent Skill) - Install (CLI): `npx skillmds@latest add rosslevinsky/plan-phase-v1` - Raw SKILL.md: https://api.skillmd.com/api/skills/rosslevinsky/plan-phase-v1/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: rosslevinsky (https://skillmd.com/u/rosslevinsky) - Updated: 2026-09-21 - Page: https://skillmd.com/skills/rosslevinsky/plan-phase-v1 --- # 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 ` 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 already refuse a v1 plan — the guard is now bidirectional.) ### Handling non-plan-init-v1 input If the file was not produced by `/plan-init-v1`, look for equivalent content: | Expected section | Acceptable equivalents | |---|---| | Goal | Objective, Summary, Overview, Problem Statement | | Success Criteria | Acceptance Criteria, Definition of Done, Tests | | Affected Areas | Scope, Files, Components, Modules | | Technical Constraints | Constraints, Requirements, Technical Notes | 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//` 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//`, 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 behaviour must include tests for that behaviour. 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 "" — N phases: Phase 1: Goal: Entry: Tasks: <3–5 bullet points of what this phase covers> Verify: Exit: Phase 2: ... 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//phase--.md` (zero-padded number, e.g. `phase-01-add-schema-migration.md`). Each phase document uses this structure: ~~~markdown # Phase : _Status: pending_ ## Goal ## Entry Criteria Before starting this phase, confirm: - [ ] - [ ] ## Tasks - [ ] - [ ] - [ ] <...> ## Tests _For logic, API endpoints, and utilities: write failing tests before implementation (TDD). For UI and wiring: write tests alongside the code._ - [ ] `` — - [ ] ## Verification Run these after completing all tasks: ```bash ``` Also verify manually: - ## 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 - [ ] - [ ] phases.md phase checkbox updated to `[x]` ## Commit ``` ``` ~~~ 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//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. ~~~markdown # Phases: _Execution tracker for [``](./)_ ## Status | Field | Value | |---|---| | Phase | Phase 1 of N — | | State | Ready to execute | | Blocker | None | | Last updated | | ## Phases - [ ] [Phase 1: ](./phase-01-.md) - [ ] [Phase 2: ](./phase-02-.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 ` to execute all phases." — its real path, not a literal `plans//plan.md`, which names nothing when the plan came from somewhere else.