Automations
Combine repeatable commands, agent judgment, and reviewable results in one workflow.
Save the work you keep setting up again. A hob automation can run a deterministic check, hand the next step to an agent, and bring back a report you can review. The steps, conversations, and results stay together in the project, so you can see what happened and run the same workflow again.
Creating an automation
Start with a command you already use: hover its
terminal block and choose Save as automation.
For a workflow with several steps, open Automations from the tool rail and
select New automation (+). The builder lets you describe the workflow to
an agent or edit its definition in the File and Configuration tabs.
Save the draft when it is ready to run.
Check and review the guestbook
Use the example project for this first workflow. The app itself needs no installation or build step; this optional syntax check requires Node.js on the machine running hob, plus a configured agent backend for the review.
The complete guestbook automation definition contains the three steps below. Download it, then run this command from a terminal in your hob example project, substituting the downloaded file's path:
hob automation init --stdin --scope private < /path/to/automation.yamlhob creates a managed bundle and prints its generated ID and exact paths. The
example deliberately omits an ID because init assigns one. Creating it does
not start a run. Choose shared instead of private if you want the bundle
in the project repository for others to use.
Check the JavaScript
- id: check
name: Check JavaScript
exec: node
args: [--check, app.js]This runs a syntax check in the project directory. A nonzero exit stops the workflow before the review. Success means the JavaScript parses; it does not prove the guestbook behaves correctly in a browser.
Ask an agent for the review
- id: review
name: Review the guestbook
agent:
prompt: |
The JavaScript syntax check passed. Read app.js and index.html.
Review note entry, character limits, and browser storage. Do not edit
the project files. Write a concise Markdown report to
${{ run.dir }}/review.md with findings and checks still needing a browser.
Distinguish code inspection from tests you actually ran.
permission-mode: ask
timeout: 10mhob opens an agent pane, sends the prompt, and waits for its turn. This manual
workflow uses Ask permissions, so be available for requests. The agent uses
your configured defaults; you can choose a backend and model explicitly in the
definition. ${{ run.dir }} gives the report a directory unique to this run.
Bring the report into view
- id: report
name: Show the review
present:
file: ${{ run.dir }}/review.mdWhen execution reaches this step, hob snapshots the report and presents it as Markdown. The file must exist: a missing report fails the step. Its captured version remains attached to run history even after temporary run files are cleaned up.
These snippets are entries in the complete definition's steps list. Use the
download above to create the workflow without assembling fragments yourself.
Run it and read the result
Open the saved automation and select Run. Each step shows its own status and duration. Expand a command to inspect its output, an agent step to read that step's conversation, or a presentation to read its captured artifact.


Succeeded describes execution, not the absence of findings. Read the report before deciding the work is ready. If a step fails, inspect its output; later steps normally remain Not run. Correct the problem and start another run. Each run is a new execution, not a resumed process.
Two ways in
Use the Automations panel for quick access to saved workflows and their Run controls. Its rows show recent status alongside the automation name.


The full Automations view contains configuration, the builder, secrets, and
run history. Open it from the panel's Overview control, or use
hob open view automations.
Run history
Open an automation's Run history to compare earlier runs and revisit their step output, agent exchanges, and captured reports. Runs also appear in project History, unless you muted the automation there. See Stats → Hours to understand which automation agent steps ran outside your working day. An earlier report remains evidence of that earlier run; rerun the workflow for a fresh result against current files.
Where automations live
Private bundles belong to this project on your machine. Shared bundles
live in the project's .hob/automations/ registry. Global bundles are
available across your projects. Use hob automation show <id-or-name> for the
exact bundle and manifest paths.
Shared bundles require human approval; Secrets, schedules, and trust explains initial review, changes, and automatic-run permission.
Declared secrets are encrypted at rest, scrubbed from automation output and history, and never expanded into agent prompts. Those protections apply to hob's automation data paths; they do not make arbitrary files or external tools inaccessible to an agent. See Secrets and trust for the allowed uses.
For agents
Create managed bundles with hob automation init, inspect their paths with
hob automation show, and check definitions with hob automation validate.
Run an existing workflow with hob automation run <id-or-name>; inspect outcomes
with hob automation runs and hob automation logs <run-id>.
Keep authored scripts and templates in the bundle and generated files in
HOB_RUN_DIR. If a shared run pauses for trust approval, tell the person what
is waiting; approval remains theirs. See the
automation CLI details.
More automation topics
| I want to… | Read next |
|---|---|
| Compose commands and steps | Building automations and script shortcuts |
| Coordinate agents and child runs | Agent result contracts, nested automations, and parallel analysis |
| Handle failures and cleanup | Retries and timeouts and finally |
| Pass data or choose a branch | Inputs, expressions, and execution environments |
| Protect credentials or run unattended | Secrets, schedules, and trust |
| Expose a result | Live status and links and artifact collection |
Next
Building automations
Add inputs, structured agent results, and explicit failure handling.
Automation reference
Find step options, inputs, expressions, secrets, and triggers.
Testing
Keep checks and failures close to the agent doing the work.
Secrets, schedules, and trust
Prepare work to run unattended and keep credentials out of prompts.