How it works

The loop has five steps: define, assign, render, sync and govern. Here is each one, with the command that does it.

Step 1
Define
Author a canonical package of instructions, skills, hooks and configs, then `wl push` it to the registry as a draft package version.
Step 2
Review and promote
The content scan runs on the new version. Once it passes, `wl publish` moves the draft to pending_review, and an approver promotes it.
Step 3
Assign
Create a workspace for a repository that pins package versions per target and provider, then `wl bind --registry <workspaceId>` connects a working directory to it.
Step 4
Render
`wl pull` resolves the packages, renders each provider’s files from the render plan, and writes the lock file and state file under .loom/.
Step 5
Sync
`wl status` and `wl diff` show each file’s ownership state. `wl sync --ci` fails the build on drift, and `wl restore` puts Workloom-owned files back.
Step 6
Govern
Trust gates hold hooks and MCP servers for approval, roles decide who can publish and approve, and the audit log records every change.

Stop copy-pasting AI config between repos

Put your instructions, skills and hooks in one registry, pin them per repository, and let CI tell you when anything drifts.