Skip to main content
Jobs are scheduled tasks defined as markdown files with YAML frontmatter. Each job lives at ~/.operator/jobs/<name>/JOB.md.

​
Job Format

---
name: daily-summary
description: Summarize today's activity
schedule: "0 9 * * *"
agent: operator
model: "anthropic/claude-sonnet-4-6"
hooks:
  prerun: scripts/check.sh
  postrun: scripts/notify.sh
enabled: true
---

Summarize the key events from the last 24 hours.
Post a one-line teaser to #general, then reply in a thread with the full summary.
The markdown body after the frontmatter is the prompt sent to the agent.

​
Model Override

Set model in the frontmatter to override the agent’s configured model for this job. Uses LiteLLM format (provider/model-name). When omitted, the agent’s model chain from operator.yaml is used. This is useful for running cheaper models on simple recurring tasks: 
---
name: link-check
schedule: "0 */6 * * *"
model: "anthropic/claude-haiku-4-5"
---

Check all links in the docs site and report any 404s.

​
Hooks

HookDescription
prerunGate script — non-zero exit skips LLM execution.
postrunReceives model output on stdin.
Hook scripts receive these environment variables:
VariableDescription
JOB_NAMEName of the job being executed
OPERATOR_AGENTAgent running the job
OPERATOR_HOMEPath to ~/.operator
OPERATOR_DBPath to the SQLite database

​
Posting Results

The agent uses send_message to post results to Slack channels. The prompt body should include posting instructions (which channels, whether to thread, etc.). If there’s nothing to post, simply don’t call send_message.

​
Job Counters

Each job tracks four counters in SQLite:
CounterIncremented when
run_countLLM actually executed (success or error)
error_countLLM ran but threw an exception
gate_countprerun hook returned non-zero (job skipped)
skip_countCron fired but the previous run was still in progress

​
Managing Jobs

Use the CLI to list, inspect, enable/disable, and manually trigger jobs.