Skip to content

creatortrack-dev-slot-lifecycle-and-teardown-button — Handoff

URL: https://mkdocs.justinsforge.com/memory/handoffs/creatortrack-dev-slot-lifecycle-and-teardown-button-2026-07-05/

Date: 2026-07-05 Owner: the agent that owns chat.justin (the CreatorTrack session launcher UI) Parent session: feat/shared-week-nav build session (wrote the teardown script + this handoff)

The Goal

chat.justin already spawns CreatorTrack parallel-dev slots. Close the loop by letting it tear one down from the same UI: a per-session "Tear down slot" button that kills the dev server, removes the git worktree, and drops the clone DB. This kills the slot-leak problem (there are already ~20+ orphaned ct_* DBs, each a full lifeos-sized copy). Optional stretch: a "Fresh data" toggle on spawn.

Current State

  • Teardown script is DONE and tested: ~/forge/scripts/forge_creatortrack_agent_teardown.sh <branch> [--keep-db]. Idempotent, faithful end-state reporting, guards refuse main/master and only ever drop ct_*. Verified: main-guard rejects, no-op branch is safe, summary reports absent/dropped/removed accurately.
  • What it does NOT yet have: a safety refusal on a dirty or unmerged worktree. It currently uses git worktree remove --force, which discards uncommitted changes. For a one-click button this MUST be gated (see Likely Approach). Recommend adding a default guard to the script (refuse if the worktree has uncommitted/unpushed work; --force to override) so the button is safe by construction.
  • Dev-data pipeline context (already fixed this session): slots clone creatortrack_dev_base, which is a snapshot of live lifeos (prod DB). The base was stale; refreshed it with forge_creatortrack_dev_base.sh --force. That script had a real bug (pg_dump aborted on app.api_connections — FORCE RLS + real OAuth creds); fixed by --exclude-table-data='app.api_connections' (also correct security: don't clone OAuth tokens into dev). Full detail in scratchpad FINDING-clone-prod-db-into-dev-slot.md.

Files You'll Work With

File Purpose
~/forge/scripts/forge_creatortrack_agent_teardown.sh The teardown action the button calls (kill server + rm worktree + drop DB)
~/forge/scripts/forge_creatortrack_agent_bootstrap.sh The spawn counterpart (branch → worktree + DB + server). Note: hard-codes ~/forge-suite; repo is now ~/creatortrack (compat symlink added). Slashed-branch names break its dev-server log path.
~/forge/scripts/forge_creatortrack_dev_base.sh Refreshes the template DB from lifeos (for a "Fresh data" toggle)
chat.justin launcher backend (you own this) New endpoint that shells out to the teardown script with the session's branch
chat.justin launcher UI (you own this) The per-session "Tear down slot" button + confirm + optional "keep DB" checkbox

Likely Approach

  1. Backend endpoint (e.g. POST /api/slots/teardown { branch, keepDb }): shells out to forge_creatortrack_agent_teardown.sh <branch> [--keep-db], returns its stdout. chat.justin already tracks each session's branch/port, so the button passes the branch it already knows.
  2. Safety gate BEFORE running (either in the endpoint or, better, add to the script as a default guard):
  3. Refuse if git -C <worktree> status --porcelain is non-empty (uncommitted work).
  4. Warn/refuse if the branch has commits not merged into main and not pushed to the backup remote (unmerged work would be lost with the worktree).
  5. --force / a UI "I know, delete anyway" to override.
  6. UI: a "Tear down" button per active slot, a confirm dialog showing exactly what will be removed (server :PORT, worktree path, DB name), and a small "Keep DB" checkbox (maps to --keep-db) for when Justin wants to inspect data after.
  7. Optional stretch — "Fresh data" toggle on spawn: since clones go stale, either (a) a nightly forge_creatortrack_dev_base.sh --force timer so every new slot starts ≤24h fresh, or (b) a per-spawn "refresh base first" checkbox. (a) is the higher-value, ~10-line systemd timer.

Don't Do

  • Don't reimplement teardown logic in JS — call the script; it owns the correct DB-name derivation and guards.
  • Don't tear down a slot with uncommitted/unmerged work without an explicit force + confirm. --force worktree removal is irreversible.
  • Don't let the button drop anything but ct_* (the script already guards this; don't bypass it).
  • Don't auto-deploy from chat.justin. Teardown ≠ deploy; keep deploy separate and manual.
  • Don't clone OAuth creds back into dev: leave the app.api_connections exclusion in forge_creatortrack_dev_base.sh.

Deliverables

  1. A working per-session "Tear down slot" button in chat.justin wired to the teardown script (with confirm + keep-DB option).
  2. A safety gate so the button can't silently destroy uncommitted/unmerged work.
  3. (Optional) Nightly base-refresh timer so new slots carry real, current data.

Done When

  • Clicking "Tear down" on a slot in chat.justin stops its server, removes its worktree, and drops its ct_* DB, and the UI reflects the slot as gone.
  • The button refuses (or hard-confirms) when the worktree is dirty or unmerged.
  • "Keep DB" leaves the ct_* DB in place while removing server + worktree.
  • No new orphaned ct_* DBs accumulate from normal spawn/teardown cycles.