tomevault

Install AI agent skills, instructions, or Tomes into any coding tool, from a quality-graded index.

npx tomevault install owner/repo

One CLI for Claude Code, Cursor, Copilot, Gemini, Windsurf, Codex, and every new tool that reads instruction files.

Why

Every AI coding tool reads a different instruction file. A setup that works in Claude Code does nothing in Cursor until you rewrite it, and most instruction files on GitHub are unrated, so you cannot tell a careful one from a careless one.

tomevault pulls from an index where files are quality-graded. Search results carry that grade, so you can judge a result before you install it, and the CLI converts what you install to whatever your tool reads.

Quick start

# Explore the index
npx tomevault search "react testing"

# See what the index knows about a repo
npx tomevault show facebook/react

# Install a Tome into your repo
npx tomevault install facebook/react

# Pull every format for your own repo
cd my-project
npx tomevault init

# See what's installed
npx tomevault inspect

# Refresh to latest
npx tomevault update

Run npx tomevault with no arguments for an interactive hub that detects your repo state and suggests next steps.

Commands

| Command | What it does | |---|---| | tomevault | Interactive hub. Detects your repo and suggests next steps | | tomevault search <query> | Search the index for Tomes, configs, and skills | | tomevault show <target> | Show what the index knows about a repo, without installing | | tomevault install <owner/repo> | Install a Tome, config, or skill from the index | | tomevault add <url> | Install skills from any .well-known/agent-skills publisher | | tomevault init | Scan your repo and pull any missing format conversions | | tomevault inspect | Show what's installed locally and what's available for this repo | | tomevault update [target] | Refresh installed content to its latest version | | tomevault convert [url] | Check if a repo is indexed, or submit it for conversion | | tomevault publish <manifest> | Validate a tome.json manifest and publish it to TomeVault |

compose, for combining multiple Tome manifests into one, is in development and not yet available in this build. Run any command with --help for its flags.

Supported platforms

| Platform | File | |---|---| | Claude Code | CLAUDE.md | | Codex / OpenAI | AGENTS.md | | GitHub Copilot | .github/copilot-instructions.md | | Cursor | .cursor/rules/*.mdc | | Gemini CLI | GEMINI.md | | Windsurf | .windsurf/rules/*.md |

See it before you install it

A plain converter just copies files. tomevault surfaces what the index knows first.

Every search result carries a quality grade where the index has scored one. To look closer, tomevault show <target> prints the index's provenance record for a repo without installing anything. When the index has a record, it shows:

• Quality: grade and score
• Security: result of the instruction-file scan
• Source: whether the content is classified as human-authored, AI-generated, or mixed
• Conversion fidelity: how many format conversions verified clean

tomevault install <target> --provenance shows the same record inline before it writes. Provenance is best-effort: when the index has no record for a repo, the install still proceeds and tells you so.

Bundle your own Tome

A Tome is a single named, versioned unit that bundles a set of skills plus repo context, so a coworker installs the whole setup once instead of copying files by hand. It is to your AI context what a Helm chart is to a cluster.

tomevault publish <manifest> validates a tome.json manifest against the same rules the server enforces, then publishes it:

# Validate without publishing
npx tomevault publish tome.json --dry-run

# Publish
npx tomevault publish tome.json

Publishing is an authenticated flow; sign in at tomevault.io first.

Write safety

Every command that touches your filesystem respects these flags:

• --dry-run: show what would change, write nothing
• --diff: show a diff and confirm before overwriting each file
• --force: overwrite existing files without prompting
• --yes: skip confirmation prompts (assume yes; useful in CI)
• --verbose: print stack traces on error

By default, existing files are never silently overwritten. Skills go to .claude/skills/<name>/SKILL.md; reference configs go to .tomevault/reference/<owner>--<repo>/.

Telemetry

Installs ping an anonymous, fire-and-forget counter so Tome pages can show accurate install counts. No personal information is ever sent. Opt out with --no-telemetry or DO_NOT_TRACK=1.

Environment variables

• DO_NOT_TRACK=1: disable anonymous install-count telemetry
• TOMEVAULT_BANNER_COMPACT=1: compact banner for scripted use

Learn more

• tomevault.io
• Every format, explained
• Report issues

MIT © TomeVault