truenas-purge-failed

Purge a TrueNAS app

This skill is the recovery path for deployments that went wrong and need a clean slate. It removes the app and its Docker volumes from TrueNAS. It does not delete host directories under /mnt/<pool>/<app-name>/ — Sven decides manually whether to wipe those.

When to use this

Trigger on phrases like:

• "Purge nginx-test"
• "Clean up the failed paperless deploy"
• "Remove that experimental stack completely"
• "Reset jellyfin so I can reinstall it"

If the user just says "delete" or "remove" without "fail" / "purge" / "completely", call delete_app(app_name, remove_volumes=False) instead — that's the gentler path that keeps volumes.

Inputs

• app_name (required).

Procedure

Step 1 — Confirm intent (mandatory)

Always show the user what will happen before doing anything:

About to purge <app_name>:
  - Stop the app (if running)
  - Delete the app definition
  - Delete all Docker volumes attached to the app
  - KEEP host directories under /mnt/<pool>/<app-name>/

Continue? (y/N)

If the user says no or is ambiguous, stop and recommend delete_app with remove_volumes=False as the gentler alternative.

Step 2 — Gather pre-state

Call get_app_details(app_name) to record the starting state. Note:

• Current status (RUNNING, STOPPED, ERROR, DEPLOYING)
• The list of volumes and their host paths
• The image name (in case the user wants to redeploy later)

If the app isn't found, stop and tell the user — there's nothing to purge.

Step 3 — Execute the purge

Call purge_app(app_name) — the MCP tool runs the whole sequence:

1. stop_app(app_name) (skipped if already stopped)
2. app.delete(app_name, {remove_volumes: true})
3. Polls the job until SUCCESS

The tool returns a structured steps[] array showing what worked and what didn't. Each entry has action and status. Don't bail on the first failure — purge_app is designed to continue past individual errors so you don't leave half-cleaned state.

Step 4 — Report what happened

Format the result for the user:

Purge of <app_name>: <ok | partial | failed>

Steps:
  ✓ stop_app          — was already STOPPED  / OK
  ✓ delete_app        — OK
  ✓ remove_volumes    — OK

Kept on host:
  /mnt/nvme-01/<app_name>/   (Sven decides manually)

If any step failed, surface it explicitly:

✗ delete_app — job 1234 FAILED: "namespace already exists"

Step 5 — Offer follow-up actions

After a successful purge, ask whether they want to:

1. Delete the host directories too (for a true clean slate) — they'd do this manually via SMB/SSH; the MCP server intentionally doesn't touch host paths.
2. Redeploy from a known-good compose file (use truenas-deploy-and-verify).
3. Nothing — leave the state as-is.

What this skill never does

• Delete host directories under /mnt/. The MCP server doesn't do this, full stop. The host filesystem is Sven's responsibility.
• Purge multiple apps in one go. Always one at a time, always with explicit confirmation for each.
• Skip the confirmation prompt, even if the user typed --force or similar — there's no --force flag in this skill.

Sequencing relative to other skills

Situation	Use this skill?	Or which other?
Deploy failed mid-way, app is in ERROR	yes	—
Just want to redeploy a working app	no	update_app or restart_app
Want to delete but keep data for migration	no	delete_app(..., remove_volumes=false)
Diagnosing why an app keeps crashing	no — diagnose first	truenas-troubleshoot