v0.18.1

Fixes from the v0.18.0 validation run

• Cancelling an action execution now actually stops it. Previously "cancel" returned success while the script kept running on the guest and the execution stayed running until it finished by itself — OpenSSH ignores signals on sessions without a PTY, and the code then waited for the remote output to end. The script now runs under a per-execution process name, cancel kills that job's whole process group on the guest (TERM, then KILL) from a second session, and the record flips to cancelled within seconds regardless. Verified live: wrapper and sleep child gone within ~3 s. (#211, #212 — closes #204)
• Deploy API returns 400 for bad requests. A missing or out-of-range cpu / memory_mb came back as 500 failed to start deployment; it is now 400 with the validation message, e.g. CPU must be between 1 and 128. Preflight blocker wording is unchanged. (#210 — closes #206)

Still open (tracked)

• #205 Proxmox: a failed post-clone config (cores/memory/net0/cloud-init) is only logged and the deployment still reports success.
• #207 Destroying a running Proxmox VM can outlast client timeouts even though it succeeds.

Upgrade

• docker pull ghcr.io/blink-zero/forgemill:v0.18.1 (or :0.18, :0, :latest)
• No database migration, no API shape changes. Drop-in replacement for v0.18.0.

Full changelog

v0.18.0...v0.18.1