cpm-next-models

Claude Code Marketplace

A Claude Code plugin marketplace whose main offering is cpm-next: a planning and building method for Claude Code, plus the mods that keep its model, effort and progress in view. It also holds smaller development tools.

cpm-next in brief

cpm-next turns an idea into working, tested code through a short chain of artefacts kept in your repository's docs/ folder: a discussion record, a product brief, a specification, and epics made of stories with acceptance criteria. Six skills cover the whole path:

Each skill declares its model and effort: Opus plans and reviews, Sonnet builds, an Opus auditor agent checks every story before it closes, a Haiku scout agent answers lookups, and status runs on Haiku. The artefact formats are defined in cpm-next/shared/artifacts.md; the plugin's own log of changes is in cpm-next/README.md.

Training material (open in a browser):

• cpm-next-onboarding.html — install it and run your first spec in an afternoon
• cpm-next-presentation.html — a slide deck introducing the method
• cpm-next-training-guide.html — the full guide: every skill, the artefacts, the mods and day-to-day practice

CPM and DPM, the earlier planning plugins, were withdrawn on 2026-10-08. Their source is in this repository's git history.

Installation

Inside Claude Code:

/plugin marketplace add ninthspace/claude-code-marketplace

The suffix is the marketplace's name, not the repository's. marketplace.json declares ninthspace-marketplace, and every installed plugin is keyed by it, so cpm-next@claude-code-marketplace resolves to nothing.

The cpm-next set

Install these at user scope (the default), so one install serves every repository:

/plugin install cpm-next@ninthspace-marketplace           # the six skills, the auditor and scout agents
/plugin install cpm-next-models@ninthspace-marketplace    # mod: holds each skill's model and effort, per-story effort
/plugin install cpm-next-progress@ninthspace-marketplace  # mod: /progress-tracker, a claude.ai tracker for a spec's epics
/plugin install whats-next@ninthspace-marketplace         # mod: /next, a live pane of the work left in the repository
/plugin install plugin-sync@ninthspace-marketplace        # mod: reloads open sessions when a plugin here is updated
/reload-plugins

Only cpm-next is required. The four mods are optional, and each adds one thing:

Mods are plugins of function hooks that run inside Claude Code. They load like any plugin, run no model calls of their own unless stated, and are switched off and on with the plugin. A session that was open before you installed one needs /reload-plugins to load it.

Per-project setup for /progress-tracker. In a repository whose spec has epics in docs/epics/, run /progress-tracker 01 (the spec's number), then ask Claude to publish the page it writes. The mod saves the tracker's link and keeps the table in step with the epic docs from then on. Details under CPM Next Progress.

Keeping up to date. The marketplace auto-updates when Claude Code starts. plugin-sync then reloads sessions that were already open. To update straight away, from any directory:

claude plugin marketplace update ninthspace-marketplace
claude plugin update cpm-next@ninthspace-marketplace      # and each other plugin that changed

Other plugins

/plugin install noteplan@ninthspace-marketplace
/plugin install php-lsp@ninthspace-marketplace
/plugin install js-simplifier@ninthspace-marketplace
/plugin install filament-mockup@ninthspace-marketplace
/plugin install generated-files@ninthspace-marketplace
/plugin install weather@ninthspace-marketplace

Available Plugins

CPM Next (v0.7.2)

Plan and build with six skills: party, plan, review, do, library, status

The skills read and write Markdown artefacts under docs/: discussions/, briefs/, architecture/, specifications/, epics/, reviews/, retros/, quick/ and library/. The artefacts are the only state; there are no progress files to clean up. A typical path:

1. /cpm-next:party to talk an idea through (optional).
2. /cpm-next:plan to write the brief, spec and epics. It reads what exists, asks its questions in one batch, and fills only the gaps. In a brownfield project it grounds every requirement in the current code.
3. /cpm-next:review before a large or risky epic (optional).
4. /cpm-next:do to build: one story (/cpm-next:do 3), one epic, or all, which runs unattended and marks a blocked story instead of stopping. Each acceptance criterion gets an Evidence line, and each story is audited before it closes.
5. /cpm-next:status whenever you return to the project.

Story effort. plan gives a story Effort: high where a mistake is costly or hard to see, and Effort: low for mechanical edits; with cpm-next-models installed, do builds each story at that effort and raises it for a fix after a failed audit. A low story marked Model: haiku is built by a Haiku subagent.

Agents: auditor (Opus, read-only) checks a finished story's evidence, diff, scope and tests; scout (Haiku, read-only) answers lookups such as where something is used or what a read-only query returns.

Quick Start:

/plugin install cpm-next@ninthspace-marketplace
/reload-plugins
/cpm-next:plan a CSV export for the bookings report

CPM Next Models (v0.3.0)

Holds each cpm-next skill's model and effort for its whole run

A Claude Code mod. A skill's model and effort frontmatter normally lasts only for the turn that invoked it. This mod keeps them for every later turn of the run, including your replies to the skill's questions, until another skill runs, you switch model or effort yourself, or /clear. It runs do all at high effort, and gives /cpm-next:do a set_story_effort tool so each story is built at its own Effort, never below high in a do all run. The footer shows what is held, for example cpm-next:do · story 3 · sonnet · low.

Commands: /cpm-models shows what is held; /cpm-models off releases it.

Quick Start:

/plugin install cpm-next-models@ninthspace-marketplace
/reload-plugins

Develop: claude plugin validate cpm-next-models and claude plugin test cpm-next-models.

CPM Next Progress (v0.3.1)

A claude.ai progress tracker for a cpm-next spec's epics, with a check that it matches the epic docs

A Claude Code mod (a plugin of function hooks). Name a spec and the mod writes a build order for its stories to a small JSON file, which you can edit; Claude then publishes a private claude.ai artifact showing that order as a table, grouped by phase. The mod compares the table with the epic docs in docs/epics/ whenever they change, and tells Claude which rows to update. Status, titles and outstanding blockers come from the epic docs; the order, phase labels, notes and open decisions come from the build-order file.

The mod only reads. It lists the artifact's rows through the ArtifactData tool, which auto mode does not ask about, and never writes to it. Claude's own ArtifactData calls do the writing, so no permission rule is needed. An earlier version wrote to the artifact itself, and auto mode refused those calls intermittently with "The server-side auto mode classifier gave no verdict for ArtifactData".

Build-order file: docs/specifications/NN-build-order.json:

{
  "title": "01-Series Build Order",
  "spec": "docs/specifications/01-spec-requirements-matrix-todos.md",
  "decisions": ["Questions still open, shown in a box above the table"],
  "phases": [
    { "label": "No visible change", "items": [
      { "epic": "01-05", "story": 1, "note": "Default value keeps today's text" },
      { "epic": "01-05", "task": "3.1" }
    ] },
    { "label": "Waiting on input", "note": "Shown beside the phase label", "items": [
      { "epic": "01-05", "story": 2, "waitingOn": "new name from the client" }
    ] }
  ]
}

An item is a story (story) or one task (task) of an epic, named by the epic's number prefix. waitingOn shows the item as Waiting on input until it is removed from the file or the item starts. artifact holds the tracker's link; the mod adds it when Claude publishes the page.

The first build order is written from the spec's epics (those whose number starts with the spec's): one phase per epic, epics ordered by their Blocked by epics, each epic's stories ordered by their Blocked by stories and then by number. Superseded and withdrawn epics and stories are left out. Edit the file afterwards to reorder, regroup, split a story into tasks or add waitingOn.

Statuses: Complete, In progress (the story says so, or one of its tasks has started), Waiting on input, Pending. A row's date moves only when its status changes. Blockers shown are the story's and epic's Blocked by items not yet Complete, earlier unfinished tasks of the same story, and any waitingOn.

How it works:

1. After any tool call that writes (Edit, Write, Bash and so on), the mod checks whether an epic doc or build-order file changed since the last check. If none did, it does nothing more.
2. If one did, it reads the tracker and works out the writes that would bring it in line. If there are none, it says nothing.
3. If there are some, it saves them to .claude/cpm-next-progress/NN-build-order.pending.json, shows you a one-line notice, and gives Claude a note with the artifact link and the writes. Claude applies them with one ArtifactData batch. More than eight writes stay in the file, and Claude hands them to a Haiku subagent, which reads the file and applies each batch, so they never enter the main conversation.

The folder .claude/cpm-next-progress/ holds generated files (the page and the pending writes); /progress-tracker 01 adds it to the project's .gitignore.

Commands:

• /progress-tracker — checks every tracker in the repository now and prints what differs, or "up to date".
• /progress-tracker 01 (or a spec file, or a build-order file) — writes docs/specifications/01-build-order.json from the spec's epics if it does not exist yet, writes the page to .claude/cpm-next-progress/, and adds that folder to .gitignore. Then ask Claude to publish the page. When it does, the mod saves the link to the build-order file and gives Claude the rows to write. init before the name still works.

Data layout (in the artifact's database, readable by anyone it is shared with, writable by editors): collection items, one document per row (01-05-s1, 01-05-t3.1); document meta/tracker with the title, spec, last update and decisions.

Quick Start:

/plugin install cpm-next-progress@ninthspace-marketplace
/reload-plugins

/progress-tracker 01

Then ask Claude to publish the page. The spec needs its epics in docs/epics/ first.

Reads: the cpm-next epic format (cpm-next/shared/artifacts.md), as What's Next does.

Develop: claude plugin validate cpm-next-progress and claude plugin test cpm-next-progress. To run the working tree, start Claude Code with --plugin-dir cpm-next-progress.

What's Next (v0.2.2)

A live pane and band showing the cpm-next work left in the current repository

A Claude Code mod (a plugin of function hooks). It reads the docs/epics/ and docs/specifications/ folders of the repository the session runs in — or the nearest folder above it that has either — and shows every story not yet Complete, in the order to build them, and every spec no epic has been planned from yet. It reads the files directly, with no model calls, so it stays current as /cpm-next:do or you edit the epics.

What it shows:

• Pane — the story in progress and its next task, every remaining story in order (doing, ready, or after Story 1 / after Epic …), and each open epic's story count. Opens by itself in a repository with work left when the terminal is at least 144 columns wide; /next opens it at any width.
• Specs without epics — each spec in docs/specifications/ that no epic was planned from, in number order. A spec counts as planned when an epic is numbered after it (03-spec-… → 03-01-epic-…) or an epic names its file in Source spec; a spec whose own Status is Complete, Superseded or Withdrawn is left out, as is a withdrawal notice (a Withdrawn or Superseded by field, or WITHDRAWN / SUPERSEDED in its title).
• Band — one line above the prompt with the next story, its next task, and how many stories are left; with no stories left, the first spec to plan.
• Next steps — an Ask Claude button (hotkey a) that asks Sonnet for a short note on what to do next, from the ordered list and the first two stories in full. The note is kept per repository across sessions and dimmed once the epics change after it was written.

Order of execution: stories already In Progress first; then the other stories of epics under way (the epic's own Status is In Progress, or one of its stories is); then everything else. Within each group, repeatedly, the ready story with the lowest epic number and story number, treating each as done before choosing the next. So working on a higher-numbered epic out of order moves it to the top once its Status says In Progress. A story is ready when everything its own Blocked by and its epic's Blocked by name is Complete; epics in docs/archive/epics/ count when resolving those dependencies. Stories whose dependencies can never be met (an unknown epic, a cycle) are listed last.

Quick Start:

/plugin install whats-next@ninthspace-marketplace
/reload-plugins

# Open the pane and print the ordered list into the conversation
/next

Reads: the cpm-next epic format (cpm-next/shared/artifacts.md) — Status, Blocked by, Story and Task fields, read case-insensitively, with Done read as Complete. Superseded and Withdrawn epics are skipped.

Develop: claude plugin validate whats-next and claude plugin test whats-next. To run the working tree instead of the installed release, start Claude Code with --plugin-dir whats-next (and uninstall the release, or both draw).

Plugin Sync (v0.1.0)

Open sessions pick up plugin updates without a manual /reload-plugins

A Claude Code mod (a plugin of function hooks). Every five minutes, and after each answer, it compares ~/.claude/plugins/installed_plugins.json with the versions of this marketplace's plugins that the session loaded. When one differs, it shows a one-line notice naming the plugins and versions, and runs /reload-plugins, which waits until the session is idle. A project-scope install is compared for sessions in that project, and the user-scope install for all others.

It does not install updates. The marketplace's auto-update does that when any new session starts, or run it yourself once, from any directory:

claude plugin marketplace update ninthspace-marketplace
claude plugin update cpm-next@ninthspace-marketplace   # each plugin that changed

Quick Start:

/plugin install plugin-sync@ninthspace-marketplace
/reload-plugins

Install it at user scope so every session loads it. A session already open needs one manual /reload-plugins to load the mod; after that it reloads itself.

Develop: claude plugin validate plugin-sync and claude plugin test plugin-sync.

NotePlan Search (v1.0.0)

Search and query NotePlan notes from Claude Code

A skill for searching NotePlan content across:

• Notes folder - Standalone notes
• Calendar folder - Daily/weekly/monthly notes
• Spaces - Team/shared notes (SQLite database)
• iCloud - If syncing via iCloud Drive

Results are sorted by most recently modified first.

Quick Start:

# Search for a term
/noteplan coffee

# List all Spaces notes
/noteplan --list --spaces

# Fetch full note by ID
/noteplan --get UUID

# Search with date filters
/noteplan meeting --after 2025-01-01

# Natural language queries
/noteplan find me everything about project planning

Key Features:

• Full-text search across all NotePlan sources
• Date filtering (--after, --before)
• JSON output for AI tools
• Direct noteplan:// URLs to open notes in the app
• Excludes @Templates, @Trash, @Archive by default (use --all to include)

Requirements:

• macOS with NotePlan 3 installed
• Python 3

View full documentation

PHP LSP (v1.0.0)

PHP semantic code intelligence for Claude Code

Adds 24 LSP tools to Claude Code for PHP files via intelephense and the lsp-mcp-server bridge.

Capabilities:

• Go-to-definition, find references, find implementations
• Hover info (type signatures, documentation)
• Code completion and signature help
• Diagnostics (errors, warnings) per file and project-wide
• Safe rename across entire codebase
• Code actions (quick fixes, refactoring)
• Call hierarchy and type hierarchy
• File analysis (imports, exports, related files)
• Document formatting

Quick Start:

# One-time setup (installs intelephense + lsp-mcp-server, configures project)
/php-lsp:setup

# Restart Claude Code — LSP auto-starts on first use

# Check everything is working
/php-lsp:status

Requirements:

• Node.js >= 18
• Git

View full documentation

JS/TS Simplifier (v1.0.0)

Simplify and improve JavaScript and TypeScript code across an entire codebase

A skill that scans all JS/TS files (or a configurable subset) and applies clarity, consistency, and maintainability improvements while preserving exact functionality. Unlike targeted simplification of recently changed files, this skill works across the whole codebase.

Three parallel analysis agents:

• Modern Syntax — ES2015+ and ES2020+ upgrades (optional chaining, nullish coalescing, async/await, const/let)
• Code Quality — Dead code removal, conditional simplification, naming improvements, error handling
• Structure & Reuse — DRY violations, module organisation, function complexity, async patterns

Quick Start:

# Simplify all JS/TS files in the project
/js-simplify

# Narrow to a specific directory
/js-simplify src/

# Only git-modified files
/js-simplify only changed

# Focus on a specific pattern
/js-simplify focus on async patterns

Key Features:

• Parallel three-agent analysis for comprehensive coverage
• Respects project conventions (CLAUDE.md, ESLint, Prettier, tsconfig)
• Configurable scope — all files, specific directories, globs, or git-changed only
• Safety-first — never changes what the code does, only how it does it
• Flags ambiguous cases for manual review rather than auto-applying

Supported File Types:

• .js, .mjs, .cjs, .jsx, .ts, .tsx

View full documentation

Filament Mockup (v1.1.0)

Build high-fidelity Filament v5 admin mockups for stakeholder sign-off

A skill that turns a product brief or spec into a single self-contained HTML file that looks pixel-accurate to a real Filament v5 admin panel — clickable enough to walk a stakeholder through every screen and flow, and throwaway by design (the real Filament build regenerates all of it natively). Mockups use the real captured Filament theme CSS and Filament's exact fi-* markup, so what stakeholders sign off on is what gets built. Not for production Filament code or customer-facing/front-end mockups.

Workflow:

• Capture — lift the compiled theme CSS and design tokens from a real Filament v5 panel
• Inventory — build an FR → screen matrix so every element traces back to a numbered functional requirement
• Build — reuse Filament's exact fi-* grammar; mark genuinely custom components with the mk- namespace
• Verify — measure with Playwright rather than eyeballing, then sign off with a coverage audit
• Hand off — write the durable routing table (docs/mockups/surface-routing.md) so the downstream builder mockup-to-filament knows which surfaces it owns (works stand-alone — no brief-to-mockups prerequisite)

Quick Start:

# Turn a brief/spec into clickable admin screens
create a Filament mockup from docs/specifications/05-spec-admin-panel.md

# Or describe it directly
mock up the admin panel for this PRD

Key Features:

• Single self-contained HTML file — opens by file://, zero environment to stand up
• Maximum fidelity to Filament's real design language (captured theme, Albert Sans, standard layouts)
• Every element traces to a functional requirement — invented UI is flagged, not silently added
• Visible mk- vs fi-* boundary distinguishes mockup scaffolding from real Filament
• Bundled scaffold, capture/verify scripts, and an fi-* grammar cheat-sheet
• Part of the mockup→build family — a producer whose output the builders consume (mockup-to-filament for Filament, mockup-to-blade for bespoke); emits a routing handoff naming the lane per surface

Requires: Node + Playwright for the capture/verify scripts (npm i -D playwright && npx playwright install chromium).

View full documentation

Generated Files (v0.1.1)

A pane listing the files Claude generated this session, each with an Open button

A Claude Code mod (a plugin of function hooks). Skills such as code-to-uml, filament-mockup and the md2docx wrapper write HTML, Office and image files, often into the session scratchpad; this pane collects them so they can be opened without finding the path.

What it records: files with the extensions .html .htm .svg .png .jpg .jpeg .gif .webp .docx .xlsx .pptx .pdf that are

• written or edited with the Write or Edit tools, or
• named in a Bash command and changed while it ran (a leading cd <dir> && sets the folder relative paths resolve against). For md2docx and pandoc runs, the .docx beside each .md named is checked too, since md2docx writes there by default.

What it shows: a "Files" pane, newest first, up to 30 files: each file's name and fo