Coordinate
Automations
Schedule agent work, create tasks from desktop, mobile, or CLI, and follow execution and recovery in your runtime.
On This Page
An automation runs an Agent Profile with a prompt, on a schedule or on demand. Definitions, individual runs, and activity are shared by the desktop, phone, and CLI through the selected runtime. Automations and orchestration tasks have separate lifecycles.
Create An Automation#
Open Activity from the application menu or with Mod+Shift+A, choose Automations in its rail, then choose New Automation. Mod is ⌘ on macOS and Ctrl on Windows and Linux. You can also start one from the Command Palette (Mod+Shift+P) with New Automation, which has no default shortcut, or from a workspace’s context menu or a section header.
The composer is a single form: a name, the prompt, and one chip each for Schedule, Destination, Agent, Reuse, and Options. Where you opened it fills in what it can. From a workspace it targets a new agent tab in that workspace; from a Git project it targets a new worktree of the project and its usual source branch. Those values are marked From Context, and you can change any of them. Opened from anywhere else, the composer asks you to choose a destination.
- Prompt supports variables including
{{automation.name}},{{run.number}},{{run.scheduledAt}},{{workspace.path}}, and{{project.name}}. Type{{to pick one. Unknown variables and malformed delimiters show as errors.Ctrl+Enter(⌘Enteron macOS) saves. - Schedule offers weekdays, daily, weekly on chosen days, every N hours, once at a date and time, or a custom cron expression, in an IANA timezone. Next Runs below the chips shows the upcoming times as you edit.
- Destination and Agent decide where and with which profile each run starts (see Destinations).
- Reuse decides where runs after the first one go (see Reuse And Unbind).
- Options shows only what applies to your choices, and its chip counts the ones you changed. A recurring schedule adds start and end dates, a run limit, and what to do with missed schedules. Every automation has overlap behavior, an inactivity timeout, retry attempts, an optional precheck command, and Notify On Success. Destinations that create a workspace add its name template, section, and tags, plus setup for a new worktree and cleanup when every run gets its own workspace.
Readiness checks the draft with the runtime shortly after you stop typing and outlines the chip that needs attention. Create And Activate saves and activates the automation; Save As Draft keeps it inactive. No repository declaration, profile permission, project approval, or revision approval is required: technical validation of the prompt, schedule, destination, and profile is the only check. Editing an Active automation keeps it Active and affects its next occurrence; runs already admitted keep their original definition. Changing the schedule recalculates from now without backfill. Closing the composer with unsaved changes asks before discarding them.
Cron Syntax#
Five fields: minute, hour, day of month, month, day of week. Fields accept *, lists (1,15), ranges (1-5), and steps (*/15); months and weekdays accept names such as JAN and MON. Both 0 and 7 mean Sunday. 0 9 * * 1-5 runs at 09:00 on weekdays. There are no seconds or @daily shortcuts. Times skipped by daylight saving do not run; repeated ones run once.
Destinations#
| Destination | What Each Run Does |
|---|---|
| New Tab In A Workspace | Opens a new agent tab with the chosen profile in an existing workspace |
| New Worktree | Creates a Git worktree and branch on this computer, from a project or as the child of a workspace, starting from the source branch you choose |
| Project Folder | Opens a workspace in the project’s registered folder, on this computer or an SSH host |
| Continue An Agent Conversation | Sends the prompt to an agent already running in a tab, listed under Running Agents |
A retried attempt of a worktree run reuses the worktree that run created.
Reuse And Unbind#
By default every run starts in a new destination (New Each Run). Reuse changes that for the runs after the first one:
- Reuse Workspace opens a new agent tab in the workspace the first run created. It applies to New Worktree and Project Folder.
- Reuse Conversation sends the prompt to the agent the first run started. If that agent is busy, the run waits without using a retry. If its terminal or agent has gone away, Alera relaunches the conversation in a new tab of the same workspace. After you take the tab over or close it, or after a run in a tab Alera opened times out or is cancelled, the next run opens a new tab in the same workspace. Continue An Agent Conversation always reuses its conversation.
A run that stays busy past the misfire grace window is skipped, or opens a parallel tab in the same workspace when overlap is set to run in parallel. Reused runs start with a short note about the previous run’s summary and error. Reuse cannot be combined with cleanup on success, since cleanup would remove what the automation returns to.
The automation detail shows where it is bound, with Open and Unbind. Unbind forgets that destination, so the next run creates a new one and binds to it. Changing the destination, turning reuse off, or removing the bound workspace clears the binding too.
Find Your Tasks#
The Automations section of Activity covers the chosen runtime and its managed local and SSH destinations. One filter bar holds the status buckets (Needs Attention, Active, Paused, Drafts, Completed, Trash), the Project, Agent Profile, and Host filters, and a search. Definitions are grouped by project or section, following the sidebar, and each group can be collapsed. Use the workspace, section, or project entry points to see associated definitions and their runs. The originating workspace controls where a definition is listed, independently of where a run executes; its section follows workspace moves.
The detail shows when and where the automation runs, its last run, the prompt, the reuse binding, and Runs and Activity tabs. Completed definitions offer cloning, running again, and trashing. Clone creates a new automation with the same settings.
Runtime And Recovery#
The runtime host owns execution. Keep it running by opening Alera or starting it with alera runtime start. Creating an automation does not change Keep Runtime Open When App Quits or Start Automations At Login. Those remain explicit application settings.
Missed schedule occurrences default to Skip, including occurrences missed while the runtime was offline. Skipped occurrences are recorded once in Activity instead of as separate runs. Missed Schedules in the options can run the latest missed occurrence or queue them. Recovery of a run already admitted is separate from missed scheduling.
When a run is interrupted, the runtime first reconciles its owner. It resumes the native agent conversation when possible, then can continue with context in the preserved workspace. The run keeps its original revision and absolute 24-hour deadline, with at most three total launches and backoff of 60 and 120 seconds. Restarting the runtime or machine does not reset that budget. An unreachable SSH owner remains reserved until its closure can be verified; a new run waits instead of launching a duplicate. A timed-out run can retain this reservation while reconciliation continues.
Watch And Control Runs#
Watch Terminal shows live output or retained output read-only. It does not send input, resize the agent’s terminal, or restart an exited process. Take Over · Stops Automatic Recovery lets you operate the agent and preserves its workspace. Older runtimes expose Open And Take Over instead of read-only observation.
Run Now runs once without changing the schedule and is available for technically valid Draft, Paused, Active, Blocked, and Completed definitions. It uses the configured precheck and overlap behavior; its menu offers Run Without Precheck, Queue After Active Run, and Run In Parallel. Run Again creates a new linked run; choose previous context in its preserved workspace or start fresh. Final runs remain unchanged.
Pause stops future scheduling while admitted runs can continue; Pause And Cancel Runs also cancels them. Resume restarts scheduling from now. Move To Trash is recoverable; Restore returns other definitions Paused and previously Completed definitions to Completed. Unbind, Move To Trash, and Pause And Cancel Runs ask for confirmation.
An optional precheck runs in the target folder; a non-zero result skips execution. The agent reports progress through alera automation heartbeat and finishes with alera automation complete, including a summary and success, failure, or blocked status. Inactivity defaults to two hours and the absolute deadline to 24 hours. Repeated scheduled failures temporarily open the circuit breaker. History and audit retention remain in application settings.
Agents And CLI#
Install the Alera Automations skill to teach agents the authoring and execution contract. The base Alera CLI skill refers to it. One command can create a definition using a partial JSON object or ordinary flags:
alera automation --json create --name 'Daily Review' --prompt 'Review changed files and summarize findings' --cron '0 9 * * 1-5' --timezone America/Mexico_City --target fresh-tab --workspace-id <workspace-id> --profile-id <profile-id> --request-key daily-review
alera automation create --file automation.json --dry-run
alera automation edit --id <automation-id> --prompt-file prompt.txt
alera automation edit --id <automation-id> --reuse conversation
alera automation unbind --id <automation-id>
alera automation run-now --id <automation-id>
alera automation --json list --workspace-id <workspace-id>
alera automation --json run-show --id <run-id>--target is fresh-tab, worktree, project-checkout, or existing-tab, and is always explicit; workspace and profile selectors can use the current terminal context. --reuse is none, workspace, or conversation, and show prints where the automation is bound. --dry-run validates without saving. A stable --request-key makes creation safe to retry after an ambiguous response. edit --file accepts changed fields and supports --expected-revision. preview-schedule displays upcoming occurrences, and clone copies a definition. Fields the composer does not show, such as the queue cap, heartbeat interval, misfire grace, retry backoff, and circuit breaker limits, can be set in the JSON definition. approve remains a hidden alias of resume for older scripts.
On The Phone#
The phone has the same composer, destinations, reuse options, filters, and detail as the desktop: buckets as chips, the other filters in a sheet, definitions grouped by project, and a binding card with Open and Unbind. Read-only automation terminals hide the composer until Take Over. Push notifications continue to report failed, blocked, and timed-out runs under Attention Required; successful runs notify only when configured. Selecting a notification opens the related automation.