Markdown
Install · 5 minutes

Install AutoVault.
One command. No daemon.

Create the local vault, admit a signed skill, scope it to the agents that need it, and run the same capability from Claude Code, Codex, or Cursor without maintaining forks.

v0.2.1 · pre-1.0 · MIT
$curl -fsSL https://autovault.sh | sh
Installer endpoint: autovault.shview source
requiresmacOS13+Linuxx64 / arm64WindowsWSL2Node20+Disk< 40 MB
~ — autovault — bash

Step 1 — Install the local vault

The installer writes ~/.autovault, installs the local CLI shim, preserves the folder as user-owned storage, and bootstraps bundled skills unless AUTOVAULT_NO_BOOTSTRAP=1 is set. Nothing runs as a background daemon; local MCP hosts spawn stdio on demand.

bash
$ curl -fsSL https://autovault.sh | sh
# macOS: also available through the tap
$ brew install autoworks-ai/tap/autovault
$ autovault skill list
Bundled skills. The installer seeds first-party bundled skills through the same validation path used by remote installs and proposals, then refreshes discovered host profiles.

Agent-assisted setup

If you want Claude Code to configure its own AutoVault bootstrap skill, give it this prompt. The hosted skill is a raw SKILL.md; the agent should fetch it, show you the behavior, install it locally only after approval, then run it.

textClaude Code prompt
Fetch https://autovault.dev/skill.md, show me what it will do, install it into ~/.claude/skills/autovault-bootstrap/SKILL.md if approved, then run /autovault-bootstrap.
Opt-in by design. The bootstrap skill stages the installer for inspection, asks before shell execution, then runs autovault doctor and autovault sync-profiles --discover.

Step 2 — Run the setup wizard

After install, autovault setup scans the vault, the bundled skills root, and any discovered native agent skill roots (~/.claude/skills, ~/.codex/skills, ~/.cursor/skills), then asks you per skill how to adopt it. Re-run any time to re-scan.

bash
$ autovault setup --review
Installed via Claude Code or another agent's shell tool? The install ran without a TTY, so the wizard was silently skipped. Open a real terminal and run autovault setup to finish onboarding. If your existing ~/.claude/skills didn't import, pick the backup adoption mode (not the augment default). See Troubleshooting for the full recovery.

Step 3 — Verify the install

One command confirms the binary, local vault folder, profile discovery, and signing key are ready before any skill enters the vault.

autovault doctor
$autovault doctor
✓ binary signed · v0.2.1
✓ ~/.autovault initialized · bundled skills indexed
✓ local keypair available · ed25519
✓ detected agents · claude-code, codex
↳ next: autovault add <source>

Step 4 — Add your first skill

Skills enter through a source adapter. Each adapter fetches from one origin and hands the raw skill to the validation gate. Whatever the source, the gate runs the same checks before admission.

bash
$ autovault add url:https://autovault.dev/skills/skill-author/SKILL.md
gate run · skill-author
↳ fetching skill-author v1.0.0
✓ yaml-repair · frontmatter clean
✓ denylist · no known bad patterns
✓ capability/behavior · declared matches observed
✓ dedup · no near match in vault
✓ sign · ed25519 provenance recorded

The vault is a folder

The final model is intentionally boring: a regular folder on disk. This is the current implementation layout: SQLite index, local signing key, source skills, source metadata, signed manifests, rendered variants, and profile links. Select a row to inspect how the folder is read.

Step 5 — Scope it to your context

By default a freshly added skill is visible only after you scope it. A caller sees a skill when it matches the agents and projects you approved, so dev-machine skills do not leak into prod and client work does not bleed across projects.

bash
$ autovault scope skill-author \
--agent claude-code,codex \
--project autovault-website \
--device $(hostname)
$ autovault sync-profiles --discover
AgentReads fromHow
Claude Code~/.autovault/profiles/claude-code/skill-authorsymlink into ~/.claude/skills
Codex~/.autovault/profiles/codex/skill-authorsymlink into ~/.codex/skills
AutoJack~/.autovault/skills/skill-author/SKILL.mdnative read of canonical source

Step 6 — Run it from your agent

The same skill is now validated, scoped, and rendered for each target agent. The skill name stays stable, while tool names are transformed to match the caller.

claude code# in Claude Code
> use skill-author to draft a new SKILL.md
✓ tool resolved: Read, Edit, Write
# extracting...
"This 24-page report covers Q1 platform metrics, with three..."

Where to next