opencode-scheduler

Run AI agents on a schedule. Set up recurring tasks that execute autonomously—even when you're away.

Schedule a daily job at 9am to search Facebook Marketplace for posters under $100 and send the top 5 deals to my Telegram

This is an OpenCode plugin that uses your OS's native scheduler (launchd on macOS, systemd on Linux, Task Scheduler on Windows), with cron fallback where native backends are unavailable.

As of v1.2.0, jobs are scoped by workdir (so different projects don't collide), and scheduled runs are supervised (no overlap + optional timeout).

Install

Add to your opencode.json:

{
  "plugin": ["opencode-scheduler"]
}

Examples

Daily deal hunting:

Schedule a daily job at 9am to search for standing desks under $300

Weekly reports:

Schedule a job every Monday at 8am to summarize my GitHub notifications

Recurring reminders:

Schedule a job every 6 hours to check if my website is up and alert me on Slack if it's down

Commands

How It Works

1. You describe what you want scheduled in natural language
2. The plugin writes a job file (scoped by workdir) and installs a timer in your OS scheduler
3. At the scheduled time, the OS scheduler calls a small supervisor script
4. The supervisor runs the job, appends logs, and updates job metadata

You can also trigger a job immediately via run_job—it runs fire-and-forget and appends to the same log file.

Jobs run from the working directory where you created them, picking up your opencode.json and MCP configurations.

Reliability Guarantees (Scheduled Runs)

• No overlap: if the previous run is still active, the next scheduled tick is skipped.
• Non-interactive by default: scheduled runs force OPENCODE_PERMISSION to deny "question" prompts, so jobs don't hang waiting for approvals.
• Optional timeout: set timeoutSeconds to hard-stop long runs (SIGTERM, then SIGKILL).

Platform Support

Windows Task Scheduler limits:

• Cron expressions that use unsupported combinations (for example, month + weekday constraints, or month-only without explicit day-of-month) return a clear error with guidance.
• Complex cron schedules may be expanded into multiple Windows tasks under \\OpenCode\\opencode-job-....
• Windows scheduled runs currently do not use the supervisor pipeline used on macOS/Linux, so no-overlap and timeout enforcement are not guaranteed by the OS integration itself.

Reference

Cron Syntax

Jobs use standard 5-field cron expressions:

┌───────────── minute (0-59)
│ ┌───────────── hour (0-23)
│ │ ┌───────────── day of month (1-31)
│ │ │ ┌───────────── month (1-12)
│ │ │ │ ┌───────────── day of week (0-6, Sunday=0)
│ │ │ │ │
* * * * *

Tools

schedule_job and update_job accept an optional timeoutSeconds (integer seconds). Use 0 (or omit) to disable.

Tools accept an optional format: "json" argument to return structured output with success, output, shouldContinue, and data.

Global Cleanup

Use cleanup_global to clean scheduler artifacts across all scopes. It always starts in dry-run mode unless you pass confirm: true.

• Dry run (safe default):

{ "confirm": false }

• Execute global cleanup of job definitions + lock files + scheduler units:

{ "confirm": true }

• Also delete logs and run history:

{ "confirm": true, "includeHistory": true }

The tool reports exactly how many artifacts were removed, grouped by location (jobs, locks, logs, runs, launchd/systemd units).

Storage

Legacy note: older versions stored jobs in ~/.config/opencode/jobs/*.json and used unscoped unit names. delete_job removes both scoped and legacy artifacts.

Working Directory

Jobs run from a specific directory to pick up MCP configs:

Schedule a daily job at 9am from /path/to/project to run my-task

By default, jobs use the directory where you created them.

Scopes

Scopes are derived from the job's workdir (normalized absolute path). This isolates job storage, logs, and OS scheduler unit names per project.

• list_jobs defaults to the current scope (your current working directory).
• Use allScopes: true to list jobs across all scopes.
• Use includeLegacy: true to include pre-v1.2.0 jobs stored in ~/.config/opencode/jobs.

Attach URL (optional)

If you have an OpenCode backend running via opencode serve or opencode web, you can set attachUrl on a job so runs use that backend:

Update the standing-desk job to use attachUrl http://localhost:4096

Project Philosophy

• This plugin is intentionally a thin wrapper: it schedules opencode run via launchd/systemd/schtasks, with cron fallback when native backends are unavailable.
• Logs are the source of truth for scheduled runs: ~/.config/opencode/logs/*.log.
• Resiliency/reporting roadmap (not implemented): PRD-resilient-execution.md.

Built-in Skill Templates

To install the built-in skill into your project (no copy/paste), open OpenCode in your repo and run:

Install the scheduled job best practices skill

This calls the plugin’s install_skill tool and writes .opencode/skill/scheduled-job-best-practices/SKILL.md.

(If you prefer, you can also say: Get skill from opencode-scheduler and add it to my skills.)

Then add @scheduled-job-best-practices at the top of scheduled job prompts.

(Manual option: use get_skill to fetch scheduled-job-best-practices and copy it into .opencode/skill/scheduled-job-best-practices/SKILL.md.)

Troubleshooting

Jobs not running?

1. Check if installed:

   • Mac: launchctl list | grep opencode
   • Linux: systemctl --user list-timers | grep opencode
   • Windows: schtasks /Query /TN "\\OpenCode\\opencode-job-*"

2. Check logs: Show logs for my-job

3. Verify the working directory has the right opencode.json with MCP configs

MCP tools not available?

Make sure the job's working directory contains an opencode.json with your MCP server configurations.

License

MIT