Remote Triggers

Remote Triggers

Create and manage remote agent triggers that execute on schedules and survive session exit.

When to Use

• Persistent schedules: Tasks that must run even after the session ends
• Cross-session automation: Recurring code reviews, automated deploys, health checks
• Scheduled agents: Run specific prompts on cron schedules in the background
• Team automation: Set up shared triggers that benefit the whole team

How It Differs from CronCreate

Rule of thumb: Use CronCreate for in-session monitoring. Use RemoteTrigger for anything that must persist. For sub-hourly or event-driven watching, use neither — see Monitor in the scheduled-tasks skill.

Tool: RemoteTrigger

Actions:

list

List all configured triggers.

RemoteTrigger: { action: "list" }

get

Get details of a specific trigger.

RemoteTrigger: { action: "get", trigger_id: "my-trigger" }

create

Create a routine. There is no top-level cron or prompt field — the schedule field is cron_expression, and the prompt is carried inside job_config.

Required: name, exactly ONE of cron_expression / run_once_at, and job_config.

{
  "action": "create",
  "body": {
    "name": "nightly-security-scan",
    "cron_expression": "0 2 * * *",
    "enabled": true,
    "job_config": {
      "ccr": {
        "environment_id": "<environment id>",
        "session_context": {
          "model": "claude-sonnet-5",
          "sources": [
            {"git_repository": {"url": "https://github.com/<org>/<repo>"}}
          ],
          "allowed_tools": ["Bash", "Read", "Write", "Edit", "Glob", "Grep"]
        },
        "events": [
          {"data": {
            "uuid": "<fresh lowercase v4 uuid>",
            "session_id": "",
            "type": "user",
            "parent_tool_use_id": null,
            "message": {"content": "Run /security-scan on the entire codebase and report findings", "role": "user"}
          }}
        ]
      }
    }
  }
}

Field notes:

• cron_expression — 5-field cron in UTC, minimum interval 1 hour
• run_once_at — RFC3339 UTC timestamp in the future; fires once, then auto-disables. Mutually exclusive with cron_expression
• job_config.ccr.environment_id — required; selects the cloud environment to run in
• job_config.ccr.events[0].data.message.content — this is where the prompt goes
• events[].data.uuid — generate a fresh lowercase v4 UUID per routine
• enabled — optional, defaults to true
• mcp_connections — optional, [{"connector_uuid": "...", "name": "...", "url": "https://..."}]; name must match [a-zA-Z0-9_-] (no dots or spaces)

The response appends the server-parsed run time and the routine's claude.ai URL — relay both to the user.

update

Update an existing routine (partial update — send only the fields that change).

RemoteTrigger: {
  action: "update",
  trigger_id: "my-trigger",
  body: { enabled: false }
}

Updatable fields: name, cron_expression, run_once_at, enabled, job_config, mcp_connections, and clear_mcp_connections (boolean, drops all MCP connections).

run

Manually trigger an immediate execution. Body is optional.

RemoteTrigger: { action: "run", trigger_id: "my-trigger" }

create_webhook_trigger

Attach an event source (for example a GitHub event) to an existing routine, so the routine fires on that event instead of a schedule. The body names the source and scope (such as a repository), the event list, a structured filter, and the routine_trigger_id to fire; the server validates the shape and rejects worker credentials. A webhook trigger has no schedule, so the response carries only the routine's claude.ai link.

delete

Not available. There is no delete action — remove routines at https://claude.ai/code/routines, or use update with enabled: false to pause one.

Execution Constraints

Routines do not run on the user's machine. Each fire spawns an isolated cloud session in Anthropic's infrastructure with its own git checkout:

• No local access — no local files, local services, or local environment variables. The cloud agent starts with zero context, so the prompt must be fully self-contained.
• The repo comes from sources — the agent sees whatever job_config.ccr.session_context.sources[].git_repository.url points at, nothing else.
• UTC only — cron_expression and run_once_at are always UTC. Convert from the user's local time and confirm the conversion with them.
• Minimum interval is 1 hour — sub-hourly crons are rejected. Use CronCreate or Monitor for anything faster.
• Tool, not curl — RemoteTrigger injects the OAuth token in-process; it is never exposed.

Use Case Patterns

Each pattern uses the full create body above; only name, cron_expression, and the prompt at job_config.ccr.events[0].data.message.content differ. All three respect the 1-hour minimum interval and are expressed in UTC.

Nightly Security Scan

• name: nightly-security-scan
• cron_expression: 0 2 * * * (02:00 UTC)
• prompt: "Run /security-scan and create a GitHub issue if critical vulnerabilities are found"

Pre-Release Quality Gate

• name: pre-release-check
• cron_expression: 0 8 * * 1-5 (08:00 UTC, weekdays)
• prompt: "Check if there are open PRs targeting the release branch. Run /quality-gate on each."

Dependency Update Monitor

• name: dep-monitor
• cron_expression: 0 9 * * 1 (09:00 UTC, Mondays)
• prompt: "Check for outdated dependencies and create a summary issue if updates are available"

Best Practices

• Use descriptive trigger names — they appear in list output
• Test a routine with action: "run" before relying on its schedule
• Avoid scheduling during peak hours — spread triggers across off-peak times
• Use update with enabled: false to pause a routine — this is the only in-tool way to stop one, since there is no delete action
• Relay the claude.ai routine URL from the create/update response — that is where execution results appear, and failed runs usually mean the prompt needs to be more self-contained

Related

• scheduled-tasks skill — Session-scoped scheduling with CronCreate (lighter weight)
• /schedule command — Higher-level wrapper for RemoteTrigger
• experiment-loop skill — Metric-driven optimization loops