Fabric Workspace Delete

/fabric-workspace-delete — Delete a Fabric Workspace (with safety hook)

Status: preview. F2 Fabric workspace lifecycle suite, 2026-05-28. Source pattern from distributions/demo-claude/.../local/_cleanup_fabric_dev.py

• _cleanup_cert_prod.py (DDF demo cleanup scripts).

⚠️ Critical safety hook encoded after the lesson in memory note "cleanup-deletes-workspace": the underlying API deletes the workspace AND all its items in one shot. The pre-flight inventory is mandatory.

What it does

Deletes a Microsoft Fabric workspace and all its items (Notebooks, Pipelines, Lakehouses, Reports, Semantic Models, …). Irreversible — there is no recycle bin.

When to use

• Tearing down a sandbox after a demo
• Cleaning up a failed onboarding attempt
• Decommissioning a deprecated environment

When NOT to use

• Anywhere near PROD without triple confirmation
• When you're not sure what's in the workspace — use /fabric-items-list first
• When the workspace contains items you cannot recreate from source — full audit via /fabric-items-list --json first, archive any non-recreateable artefact

Prerequisites

• FabricConnector credentials
• The identity must be Admin on the target workspace (Member / Contributor is not enough to delete)

Usage

/fabric-workspace-delete --workspace-id <id>                # interactive
/fabric-workspace-delete --workspace-id <id> --confirm-name "AcmeSales_DEV"  # name double-check
/fabric-workspace-delete --workspace-id <id> --confirm-name "AcmeSales_DEV" --yes  # non-interactive (CI)
/fabric-workspace-delete --workspace-id <id> --dry-run      # inventory only, no DELETE

Safety hook (mandatory, encoded by design)

The skill always performs these steps before issuing DELETE:

1. Pre-flight inventory: call connector.client.list_items(workspace_id) to enumerate ALL items in the workspace. Group by type. Render counts per type (Notebook, Pipeline, Lakehouse, Report, SemanticModel, …).
2. Name double-check: require --confirm-name "<exact workspace name>". The skill fetches the workspace via get_workspace(id) and compares the displayName field; mismatch → refuse.
3. Echo plan: "About to delete workspace <name> (id <id>) and its N items: . This is irreversible. Type the workspace name again to confirm: ___".
4. Final confirmation: operator types name a second time (unless --yes). The skill compares character-by-character. Any mismatch → abort.

--dry-run performs steps 1-3 without 4 (no DELETE issued).

Pipeline summary

1. Load project + credentials, build connector
2. Pre-flight inventory + name double-check (see Safety hook above)
3. Echo plan + final confirmation prompt (skip only with --yes)
4. connector.client.delete_workspace(workspace_id) → DELETE
5. Confirm 200/204 response
6. ops.log: WORKSPACE-DELETE | <env or -> | fabric: name=<name> id=<id> items=<n> | ok (always log even on dry-run, with detail dry-run)
7. Suggest cleanup of stale overlays/*.yaml entries that referenced the deleted workspace_id

Preview tracking — known unknowns

1. No recovery: there is no soft-delete or undo. Repeat the warning.
2. Capacity unbinding: deleting a workspace unbinds it from its capacity. No special action required, but be aware of capacity billing implications.
3. Linked datasets: reports and semantic models referenced by external workspaces (cross-workspace lineage) will silently break after delete. V1 does not detect this — future enhancement: --check-external-refs.
4. Idempotent re-runs: if the workspace is already deleted, the API returns 404. V1 surfaces as fail; should consider as ok (already in desired state) — defer to first real use.

Status — promotion to stable

1. 3+ deletes executed correctly across different workspaces, with the inventory + name-confirmation flow followed and validated.
2. At least 1 case where the name-confirmation aborted a wrong-workspace delete attempt.
3. ops.log entries reviewed and confirmed to capture workspace name + id + item count for post-hoc audit.
4. No silent destructive paths — refuse on any uncertainty.

ARGUMENTS: $ARGUMENTS