Setup Oz Automations

Set up Oz automations

Create the Oz environment and schedules that run enrich-notes-loop-cloud nightly and refresh-wiki-cloud weekly against the user's Obsidian Sync vault.

0. Prerequisites

Check whether the Oz platform skill is installed — it teaches the oz CLI. If it's absent, install it:

npx skills add warpdotdev/oz-agent-skill

Read that skill before running any oz commands.

Then gather from the user:

• Obsidian Sync vault name — e.g. Main. They need an Obsidian Sync vault pointed at their notes directory, and the headless CLI auth token for it.
• Schedule times — default to nightly enrichment and weekly wiki refresh if they have no preference.

The skills are pulled straight from bholmesdev/llm-knowledge-base-skills — no repo of their own needed. Only ask about a repo if the user says they've forked the skills to tweak them; then substitute their owner/name everywhere bholmesdev/llm-knowledge-base-skills appears below.

Make sure they're logged in (oz login). If they have no account, send them to oz.dev → "start an agent".

1. Store the Obsidian auth token as a secret

oz secret create --personal OBSIDIAN_AUTH_TOKEN

The user pastes their token when prompted — never ask them to put it in the conversation.

2. Create the environment

The setup commands install the Obsidian headless CLI and sync the vault to ~/vault on boot:

oz environment create \
  --personal \
  --name "Knowledge base automation" \
  --docker-image "warpdotdev/dev-base:latest-agents" \
  --repo "bholmesdev/llm-knowledge-base-skills" \
  --setup-command 'npm install -g obsidian-headless' \
  --setup-command 'mkdir -p ~/.obsidian-headless && printf "%s" "$OBSIDIAN_AUTH_TOKEN" > ~/.obsidian-headless/auth_token' \
  --setup-command 'mkdir -p ~/vault && cd ~/vault && ob sync-setup --vault VAULT_NAME && ob sync'

Get the environment ID with oz environment list for the next step.

3. Create the schedules

One nightly run for note enrichment, one weekly for wikis. Each points at its cloud skill, running on Kimi k2.6 — an open-weight model that's token-efficient and plenty capable for this work (swap --model if the user prefers something else; oz model list shows the options):

oz schedule create \
  --personal \
  --name "Enrich notes loop" \
  --cron "0 8 * * *" \
  --environment "ENV_ID" \
  --model kimi-k26-fireworks \
  --skill "bholmesdev/llm-knowledge-base-skills:enrich-notes-loop-cloud" \
  --prompt 'Run the enrich-notes-loop-cloud skill.'

oz schedule create \
  --personal \
  --name "Refresh wikis" \
  --cron "0 8 * * 1" \
  --environment "ENV_ID" \
  --model kimi-k26-fireworks \
  --skill "bholmesdev/llm-knowledge-base-skills:refresh-wiki-cloud" \
  --prompt 'Run the refresh-wiki-cloud skill.'

4. Wrap up

Tell the user:

• Both schedules are live, with names, cron times, and schedule IDs.
• They can pause one anytime with oz schedule pause SCHEDULE_ID.
• They can view runs and trigger a schedule manually from the online portal: https://oz.warp.dev/schedules. Output that link.

Offer to pause the schedules initially if they'd rather trigger a test run from the portal first.