Skip to main content

​
Deployment

One customer engagement: “Acme Logistics”, “Bolt Health”. Everything below belongs to one deployment, and your coding agent works on one deployment at a time. A workspace is your team; it holds all your deployments.

​
Sources

What the customer said or wrote: call transcripts, tickets, emails, docs. Upload files, paste text, have your agent add files from your machine, or connect tools like Gmail, Slack, Linear or Zendesk once for the workspace and pick what feeds each deployment. Find them in the deployment’s Sources, inside the Customer tab. Your agent searches them before assuming anything.

​
Cited plan

What you’re building for the customer: the goal, what’s in scope (with priority and status), what’s out of scope, and open questions. It’s the deployment’s Build tab; each in-scope item is a work item. The Customer tab holds their people, systems, rules and the words they use. Every item quotes the customer word for word, with a link to where they said it: [Kickoff call, 00:14:22]. A quote that isn’t really in the source is thrown away, so nothing in the plan is made up. The MCP tools call the plan and customer profile together the harness (as in generate_harness).

​
Plan change

The plan only changes when a person says so. When new material shows something new or different, OpenScaffold (or your agent) proposes a plan change: what changes, why, and the quotes behind it. You approve, edit or reject it. Every approved change is a new version; old versions stay, and rolling back adds another.

​
Review

An independent check of a change before it ships: does it do the work items it claims, and does it keep the customer’s rules? Your coding agent asks for one; OpenScaffold’s reviewer does it. See The review.

​
Needs you

What’s waiting on you, at the top of the tab it belongs to, shown only when there’s something. Customer: plan changes from new customer material, questions to ask the customer, sources that failed or need reconnecting. Build: plan changes from coding-agent sessions, reviews that need changes. Playbooks: a suggested playbook to use, new entries waiting for an admin. The orange number on a tab is how many.

​
Activity

What happened: what was added, what changed, and what each coding-agent session did. Agents report their work back at the end of a session, so you see it without asking. Open it from the deployment’s ⋯ menu.

​
Playbooks

Your team’s recipes for a kind of work (“Computer-use agents”, “Evals”): setup steps, environments, domain knowledge, caveats, links. They belong to the workspace and are shared by every deployment that uses them. They never hold customer data: every entry is checked for customer names, people, domains, systems, their terms and copied passages, and refused if it has any. New entries from agents wait for an admin.