Claude Storehouse 1.4.0

This release makes the dashboard feel faster when it opens. The page shows its layout at once, and usage now loads at the same time as the rest of your setup.

Added

• While your setup loads, the page shows a skeleton in the shape of the page: the usage band, the tiles, and the type cards on the Dashboard, and rows on other pages.
• While usage loads, the usage band shows its own skeleton. The rest of the Dashboard is ready before it.
• The shimmer stops when your system asks for reduced motion. Screen readers hear that the page is loading.

Performance

• The Dashboard, Skills, Agents, and MCP pages start to read usage when the page opens. Before, usage started only after the inventory arrived, so the two waits added up.

Changed

• The browser code is split into small modules, and the stylesheet into 13 files. The pages look and work the same.
• Development: tests for each browser module, a UI structure skill and reviewer agent, and an architecture overview.

Claude Storehouse 1.3.0

This release helps you see what your Claude Code setup costs and what it really uses. New pages show the tokens each item adds to every session, find config that is broken, and show the settings value that wins in this folder. Skills, agents, and MCP servers now show how often Claude used them in the last 30 days.

Added

Use per item

• Skills, agents, and MCP servers show "Used N× in 30 d" and the time of the last use. The counts come from the tool calls in your session files.
• An "Unused 30 d" filter and a sort by use: Most used, Least used, or Name.
• A banner lists the items with no use in 30 days and the tokens they add to every session. "Switch off N" switches them off after a confirm dialog. Each switch gets its own Activity entry with Undo.
• An item is marked unused only when your session files cover 30 days and the item file is older than 30 days. Skills with disable-model-invocation: true are never marked, because Claude cannot call them.

Context page

• Estimates the tokens that each CLAUDE.md file, rule file, skill, agent, and MCP server adds to every session, with a bar by type and filter chips.
• The Off chip shows what each switched-off item saves. Rule files with paths: show their cost on use.
• Skill and agent rows show a small token chip.

Doctor page

• Lists config that is broken or does not do what it says: invalid JSON, hook paths and MCP commands that do not exist, unset ${VAR} values in MCP servers, unknown plugins in enabledPlugins, skillOverrides that do nothing, duplicate names, missing descriptions, broken links, and long CLAUDE.md files.
• Rows are grouped by Error, Warning, and Info. The sidebar count shows errors and warnings.
• Doctor only reads. It never writes a file or runs a command.

Settings

• The Settings page has a numbered file list (Global, Project, Project local) and one file at a time, with Form and Raw JSON tabs.
• The form gives each field a plain name, help text, and its JSON key. It adds permissions.defaultMode.
• The new Effective view shows the value Claude Code uses for each key, the file it comes from, and the values it overrides. A form field that a higher file overrides says so.

Dashboard

• A This month / All time switch for the usage band. The browser remembers your choice.

Changed

• Activity is now its own page after CLAUDE.md, so the sidebar does not scroll. The sidebar has a footer with links.
• The Settings form no longer has an autoMode switch. autoMode holds classifier rules, not an on/off value.
• The usage scanner now also reads the tool name and the skill, agent, or MCP server name of each tool call. It never keeps prompts, arguments, or message text.
• The website and the live demo show all the new pages, with new screenshots.

Fixed

• Doctor no longer warns about plugin items that share a name with a local item, or about skillOverrides keys for bundled skills.
• Doctor reads a hook path only up to the first shell operator, and skips the built-in plugin marketplaces.
• The rule file scan skips links that point outside the home and project folders.
• The usage scan skips a symlinked subagents folder.

Claude Storehouse 1.2.0

This release adds a Duplicates page that finds skills, agents, and MCP servers that use the same name, so you can see which ones Claude Code loads and keep only the one you want.

Added

• Duplicates page: groups skills, agents, and MCP servers of one type that have the same name, across global, project, and plugin scopes. A group shows "Conflict" when more than one item in it is on, and "OK" when one or none is on. The sidebar count shows the groups that need action.
• "Keep only this" on each item in a group switches off the other items that have their own switch. Plugin items show the plugin that controls them.
• A group notes when its items have the same description, which usually means that one item is a copy of the other.
• The website and the live demo show the new page.

Changed

• The open dependency is now version 11. Node 20 or later is still required.
• Development: Node 24 is pinned for development, the dev tools and GitHub Actions are updated, the code passes the ESLint 10 rules, and the npm package has more keywords.

Claude Storehouse 1.1.0

This release adds a usage band that shows what your Claude Code use would cost at API prices, a Mods page, a new Activity panel, and a website with a live demo.

Added

• Dashboard usage band: API-equal cost of your Claude Code use, by model, with total tokens, sessions, projects, and how much the prompt cache saved. It reads only token counts from the session files in ~/.claude/projects, never message text. A model with no known price counts tokens but no cost, and the band names it.
• Mods page: installed mods (plugins that change Claude Code itself) and the mods in ~/.claude/dev-mods. Installed mods also stay on the Plugins page with a "Mod" flag. Mods in development are read-only.
• App icon: a terminal prompt with the Claude spark as the cursor, in the browser tab.
• Website at https://sumonta056.github.io/claude-storehouse/ with a live demo. The demo runs the real dashboard in your browser on a sample setup, so you can try switches, favorites, edits, and undo before you install.

Changed

• Activity panel: an icon for each kind of change, the time of each change, and an empty state that says what shows there.
• Hooks and MCP servers pages: a new layout.
• Favorites: the star opens a category picker. Search for a category or type a new name to create it.
• Drawer: a more compact header.
• New README with a feature tour and screenshots of every main page.

Fixed

• Skill and agent descriptions that span more than one line in the frontmatter now show in full.

Claude Storehouse 1.0.0

First stable release. Claude Storehouse is a local web dashboard for your Claude Code setup. Run npx claude-storehouse in any folder to see, edit, switch on or off, and favorite every skill, agent, plugin, hook, MCP server, settings file, and CLAUDE.md, for your global ~/.claude and for the current project.

Added

Browse

• One page with Dashboard, Favorites, Skills, Agents, Plugins, Hooks, MCP servers, Settings, and CLAUDE.md views.
• Scope filter (All, Global, Project) and search with /.
• Dashboard with counts, favorites, recently changed items, and the config files this project has.
• MCP servers from ~/.claude.json, .mcp.json, and plugins, with env and header values masked.
• Global-only mode when you run it from your home folder.

Switch on and off

• Plugins through enabledPlugins, and global or project skills through skillOverrides, the keys Claude Code itself reads.
• MCP servers through the .mcp.json lists or the per-project disabledMcpServers list in ~/.claude.json.
• Hooks are parked in ~/.claude/storehouse.json, and agents are renamed to .md.disabled.
• Plugin skills, agents, and hooks follow their plugin and show "via plugin".
• Project switches write .claude/settings.local.json, never the shared settings.json.

Edit and manage

• Detail drawer to read, edit, and save any file you own. Plugin cache files are read-only.
• Add and remove hooks and MCP servers.
• Create a missing CLAUDE.md, CLAUDE.local.md, .claude/settings.json, .claude/settings.local.json, or .mcp.json.
• Fork a plugin skill into ~/.claude/skills.
• Uninstall a plugin through claude plugin uninstall, with the right scope.
• Open a file in your editor or show it in Finder.
• Delete moves a file to ~/.claude/backups/storehouse/trash/.

Favorites

• Star skills, agents, and plugins, and drag them into your own order.
• Group favorites in your own categories. Rename and delete a category, and drag items between categories.
• A favorite stays listed when you switch it off, and shows "missing" when its file is gone.

Undo and backups

• Activity log with undo for every toggle, favorite, and category change.
• Restore the latest save of any file from its backup.

CLI

• Flags --dir, --home, --port, --no-open, --help, and --version.

Security

• Every write makes a backup, writes a temp file, and renames it. A file that does not parse is never overwritten.
• The server listens on 127.0.0.1 only, rejects a foreign Host header, and needs the random token from the start URL on every API call.
• Reads and writes are limited to ~/.claude, ~/.claude.json, and the project folder.

Changed

• New README with screenshots, MIT license, and full npm package metadata.
• Development: ESLint, Prettier, Husky, and commitlint; CI on Node 20 and 22 on Ubuntu and macOS; tag-driven npm publish with provenance.