AutoVault Bootstrap

Use this skill when the user asks to install, configure, repair, or verify AutoVault for the current agent environment.

Safety Rules

• Do not execute the installer until the user explicitly approves the reviewed script.
• Do not stream a network response directly into a shell.
• Keep every downloaded installer in a temporary file so the user can inspect it.
• If a command needs broader shell, filesystem, or network access than this skill declares, stop and ask for approval with the exact command.
• Never print secrets, tokens, private keys, or full environment dumps.

Workflow

1. Detect the host and current state:

   uname -a
   command -v autovault

2. If autovault already exists, run verification first:

   autovault doctor
   autovault sync-profiles --discover

   Report the result and stop unless verification fails.

3. If AutoVault is missing, stage the installer for review:

   tmpdir="$(mktemp -d "${TMPDIR:-/tmp}/autovault-installer.XXXXXX")"
   installer="$tmpdir/install.sh"
   curl -fsSL https://raw.githubusercontent.com/autoworks-ai/autovault/main/scripts/install.sh -o "$installer"
   wc -c "$installer"
   sed -n '1,160p' "$installer"

4. Summarize what the staged script appears to do. Include install location, profile changes, network calls, and any commands that modify the filesystem.

5. Ask the user for explicit approval before running:

   sh "$installer"

6. After an approved install, verify and sync:

   autovault doctor
   autovault sync-profiles --discover

7. If anything fails, explain the blocked step, likely permission or network cause, and the exact command the user can run manually.

Expected Outcome

The current machine should have a working autovault command, a healthy ~/.autovault folder, and discovered agent profiles refreshed through AutoVault.