Read every project from the CLI and HTTP

The Web app's All projects mode reads the same aggregates any script can. This page walks through the CLI first, then the HTTP/UDS form, with the output you should expect.

Expect about ten minutes.

Before you startCopy Anchor Link

• The daemon is v0.4.0 or later, and you run commands as the operator (not from inside an agent session). Agent sessions are refused with agent_scope_denied unless their agent holds the unrestricted grant.
• Two projects are registered. The examples use checkout-api and marketing-site.

Step 1: One flag on every listCopy Anchor Link

Add --all-workspaces to a list command. It reads every registered project plus the User scope, for the profile you are acting as. It cannot be combined with --workspace.

compozy agent list --all-workspaces

agents
======
PROJECT         NAME              PROVIDER
-------         ----              --------
User            code_implementer
marketing-site  code_implementer
checkout-api    code_implementer
User            general

Output trimmed. The PROJECT column names the project, User for the User scope, or ? when a row's project is unknown. Session lists retain their existing Workspace and other columns and add Project under --all-workspaces. The same flag works on:

compozy session list --all-workspaces
compozy task list --all-workspaces
compozy loop list --all-workspaces
compozy loop runs --all-workspaces
compozy loop requests --all-workspaces
compozy automation jobs --all-workspaces
compozy automation triggers --all-workspaces
compozy automation runs --all-workspaces
compozy automation suggestions --all-workspaces
compozy worktree list --all-workspaces
compozy terminal list --all-workspaces

Step 2: Find every running GoalCopy Anchor Link

Goals are Loop runs started from a session:

compozy loop runs --all-workspaces --origin session --status running -o json

JSON output keeps workspace_id on each run; User runs have it empty. Join project names from compozy workspace list -o json (or GET /api/workspaces).

Step 3: Use HTTP or UDSCopy Anchor Link

Every aggregate takes all_workspaces=true. Use the HTTP address from compozy status:

curl -s "http://localhost:2123/api/loop-runs?all_workspaces=true&origin=session&status=running"

New aggregate routes need the parameter. Without it you get:

{ "error": "choose workspace_id or all_workspaces=true", "code": "workspace_breadth_invalid" }

Step 4: Narrow to some projectsCopy Anchor Link

Repeat workspace_ids to keep only some projects. @user keeps User rows:

curl -s "http://localhost:2123/api/tasks?all_workspaces=true&workspace_ids=ws_a&workspace_ids=@user"

workspace_ids only works together with all_workspaces=true, and all_workspaces=true cannot be combined with a single-project selector:

{
  "error": "all_workspaces conflicts with a workspace selector",
  "code": "workspace_breadth_invalid"
}

global is not a project name on data routes:

{ "error": "global is reserved without a migration alias", "code": "workspace_reserved" }

Step 5: Page through resultsCopy Anchor Link

Paged aggregates return a cursor. Send it back with the same breadth, filters, and profile. Changing any of them invalidates the cursor; start again from the first page.

Rules worth knowingCopy Anchor Link

• Profiles. all_workspaces never widens the profile. Add all_profiles=true (operator) to read every profile. Loop definitions and agents resolve for one concrete profile and reject all_profiles=true.
• Loop runs are newest first (created_at DESC, id DESC). order=operational needs a single workspace.
• Loop definitions return entries, per-project errors, and truncated (200 per project, 2,000 total). A definition from your user or profile layer appears once on a User row; it also appears on a project row when it ran in that project in the last 30 days.
• Automation runs carry workspace_id (the project of the job or trigger that started them) and attribution: recorded, derived for older runs whose job still exists, or unknown. Unknown runs never match a workspace_ids filter.
• Worktrees come from cached state; --refresh cannot accompany --all-workspaces.
• Terminals aggregate only on the local listener.
• App tokens are refused on the new aggregate routes. On routes they could already read (/api/tasks, automation jobs, triggers, and runs, /api/agents/catalog), nothing changes.
• The needs-you bell (GET /api/notifications/attention) is always every project and profile. User items are labeled "User".