| name | Workspace Safety |
| description | Use when the user wants backup, export, checkpoint, or before bulk deletes, rm -rf, or wide refactors. |
| version | 1.0.0 |
| category | bundled |
| primary-tools | ["write_file","delete_file","move_file"] |
| tags | ["backup","export","checkpoint","safety","destructive","rollback","isolate"] |
| triggers | ["backup","export profile","checkpoint","rollback","before delete","rm -rf","wipe workspace","risky refactor","save state","delete_file","move_file"] |
Tool contract (read first)
| Need | Action |
|---|
| Conversation checkpoint | UI slash /checkpoint [name] — not a built-in tool |
| Restore checkpoint | /rollback — UI |
| Profile portability | Workspaces tab Export/Import JSON — UI |
| Isolate experiment | make_dir under work/<slug>/ — project-scaffold |
| Inventory before delete | browse_workspace (action=list / action=tree) |
| Destructive file ops | delete_file, move_file — confirm scope first |
| Risky shell | run_shell — prefer work/ subtree — browser-runtime-map |
| Multi-step irreversible plan | task-execution inserts checkpoint todo as step 0 |
Non-negotiable: No wide delete/refactor without checkpoint or export. Redact secrets before sharing exported JSON — see artifact-delivery.
Canonical scope
This skill owns backup, checkpoint, export, and risk isolation before destructive or high-value work. For where to put new isolated trees (projects/ vs work/), see project-scaffold—use both when starting a new subtree that could still go wrong (scaffold the path, then checkpoint/export if it matters).
When to Use
- Before bulk deletes, wide
run_shell refactors, or experiments that could break the tree.
- User wants portability across machines or browser profiles (export/import).
- Recovering from a bad turn without losing everything (
/checkpoint, /rollback).
- "Delete everything", mass file removal, or irreversible migration.
Relation to other skills
- Path layout:
project-scaffold. Delivery redaction: artifact-delivery. Multi-step runs: task-execution.
Checklist
- Named history checkpoint — If the UI exposes slash commands (
/checkpoint [name] per README), save conversation state before big moves; /rollback lists or restores.
- Profile / workspace export — Workspaces tab: Export profile snapshot to JSON; Import later. Do this before risky migration or "let's wipe and retry."
- Isolate experiments — New disposable trees under
work/<purpose-slug>/; durable demos under projects/<slug>/. Call skill (action=view) project-scaffold when layout is unclear.
- Secrets — Never export and share profile JSON without redacting tokens; see
artifact-delivery.
- Destructive tools — Confirm scope before
delete_file, wide patches, or shell that removes files.
- Inside multi-step runs —
task-execution inserts a checkpoint todo as step 0 when the approved plan contains an irreversible step; do not skip it.
Pitfalls
- Assuming browser storage is backed up — it is local; export if the work matters.
- Running irreversible shell on the main repo root when a
work/ subtree would suffice.
Anti-patterns
- "Let's delete half the workspace" without checkpoint or export.
- Putting irreplaceable state only in session memory — export facts/skills separately if needed for portability.
