wipelisted

# Wipe — destructive reset of Atelier state This is the nuclear option for getting back to a clean slate. The wrapped script (`scripts/wipe.sh` → `scripts/wipe.py`) handles the full cleanup including state outside the workspace dir (git worktree registry, Claude SDK transcript cache). ## What gets wiped | Surface | Wiped? | Notes | |---|---|---| | `~/Atelier/works/<slug>/` (works DB rows + on-disk dirs) | ✅ | Cascades to agents, artifacts, handoffs, transcripts. | | `~/Atelier/projects/<slug>/` (project DB rows + on-disk dirs) | ✅ | | | Per-agent git worktrees | ✅ | `git worktree remove → --force → rmtree + prune` per agent. | | Legacy `atelier/<work>/<agent>` branches in source repos | ✅ | Best-effort delete; custom user-named branches are NOT touched. | | `~/.claude/projects/<munged-workdir>/` (Claude SDK session JSONL) | ✅ | Per agent's workdir, deleted after the work-dir rmtree. | | Connections (DB rows + keychain entries) | ❌ — preserved | Re-running the wipe doesn't make you re-enter Jira/Sentry/Honeycomb tokens. | | `schema_version` | ❌ — preserved | The DB stays at the current schema version; no re-migration needed on next start. | | Amp threads on Sourcegraph's servers | ❌ — out of scope | Server-side; would need `amp threads delete` per session. | ## Steps ### 1. Confirm the backend is stopped The supervisor's in-flight writes will race the wipe's deletes if it's running. Ask the user: > Is the Atelier backend stopped? If it's running, kill it first (Ctrl+C i