v0.15.0

Added

• The official Docker image now includes bash and tzdata on both the alpine and debian variants.
• New -docker image tags (latest-docker, latest-alpine-docker, latest-debian-docker, and the versioned equivalents) add a Docker CLI with the Compose plugin. For tasks that shell out to docker/docker compose from inside the RunWisp container itself — e.g. to manage sibling containers over a mounted Docker socket. See Docker.

Changed

• The run detail header now leads with the outcome — a verdict line (succeeded in 933ms) over one fact line with the time, date, trigger, queue wait, retries and parameters. Queue wait, retries and parameters appear only when they apply, and a failing run colours the verdict and tints the header.

Fixed

• npm install -g runwisp, bun add -g runwisp, and npx runwisp now install and run the runwisp command. The npm package now ships a small launcher that execs the bundled platform binary directly (no resident wrapper process), so every documented install path works — not just bunx.
• A task removed by runwisp reload now finalizes any run still waiting in its queue, recording it with a terminal status so the run history and API never show it as perpetually pending.
• min_free_space keeps enforcing free-disk headroom across log rotations, so the safeguard stays active for the full duration of a long-running task whose output rotates. See min_free_space.
• Log line numbers stay continuous after a run’s output rotates more than once, keeping log-search results and paging anchors pointing at the right lines.
• The data directory is always created with 0700 permissions, including when it already exists (for example a pre-created Docker volume), so its contents stay private to the daemon’s user.
• Notification log excerpts are always valid UTF-8 — trimming a captured-output tail to fit a message no longer cuts through a multi-byte character.
• Release tarballs now package runwisp as an executable (mode 0755).
• CPU and memory stats now report real host values on macOS. The system-resource collector only understood Linux /proc, so on macOS the dashboard, TUI, /api/system and /metrics showed the daemon’s own Go heap as “total memory” with CPU stuck at 0%. macOS now reads host RAM and load average directly (via sysctl and a mach VM-statistics call), matching the Linux figures.
• The TUI’s shutdown command now targets the daemon it’s connected to via --socket, and surfaces an error if the shutdown fails.
• The web UI no longer shows connection status twice. The header’s connection pip and timezone badge are gone; the sidebar footer is now the one place for both, with the timezone’s source (config vs. detected) moved into a tooltip instead of sitting in the label.
• The dashboard now seeds /api/system and /api/info once per session instead of re-fetching them on every auth state change; live values already stream over /api/stream.
• Multiple open dashboard tabs no longer stall each other’s live updates. The tab holding the shared event stream now hands it off when it’s backgrounded, so a visible tab always keeps the connection.
• A cron schedule with more than one tick per hour (e.g. 0,30 2 * * *) no longer re-fires on the DST fall-back day. The dedup only remembered the single most recent firing, so the second tick of the rewound hour slipped past as a “new” one.
• A restart shortly after a large missed-run backlog no longer re-alerts on the same gap. When the backlog was capped by max_catch_up_runs, the next restart re-counted from the same anchor and reported the gap again.
• overlap = "queue" no longer lets a fresh trigger start ahead of runs already waiting. A trigger arriving in the brief window after a slot frees but before the queue drains now joins the back of the queue instead of racing ahead.
• overlap = "terminate" no longer lets active runs exceed max_concurrent under rapid triggers. Repeated triggers could re-cancel the same already-dying run instead of picking a fresh victim, letting the active set grow unbounded.
• runwisp reload now gives newly added or rescheduled jittered tasks their start-spread immediately, instead of firing at the raw tick until the next restart.
• Fixed a race where a task’s live run state could be read while still being mutated, surfacing torn or stale values in GetActiveRuns callers.
• Fixed a race that let a single-use login token or launch ticket be redeemed twice when two requests presented it at the same moment.
• Coalesced notifications now keep the first occurrence’s timestamp and run ID instead of overwriting them with the latest occurrence’s.
• A dropped control-plane heartbeat no longer tears down an otherwise healthy session. Liveness is already governed by the watchdog; a single failed send now logs and continues instead of forcing a reconnect.
• A 401/403 from the control-plane sync endpoint is no longer treated as a hard authentication failure. The WebSocket handshake is the actual auth boundary, so these now retry with backoff like any other transient error.
• runwisp promote on a live cron-sourced task now comments out the crontab line it came from, instead of leaving it live. Previously a promoted task lost its “held” status the moment it became native, so if a system cron daemon was still running unmasked, both it and RunWisp would fire the job. The line is commented, not deleted — it stays visible with a note pointing at your runwisp.toml. Promote refuses, writing nothing, if that line has changed or gone missing since the config was loaded. See CLI.