cron-scripts 🕐#

Run bash scripts on a schedule — no LLM, no agent turns, zero model cost.

This is a practical workaround for the missing payload.kind: "script" in OpenClaw cron ( issue #80793, issue #18160). It uses a gateway startup hook to load a directory of shell scripts and schedule them with the same croner library OpenClaw already bundles — no extra dependencies.

When to use this (not LLM crons)#

Use script cronUse LLM cronHealth checks, HTTP pingsSummarising logsFile/lock cleanupDeciding what to reportFixed Telegram/webhook notificationsWriting weekly digestsGit pulls, backupsAny task needing reasoningWatching directories for changesClassifying content

If a task doesn't need a brain, it shouldn't spin up one.

Installation#

1. Install the gateway hook#

Copy hooks/cron-scripts-loader/handler.ts from this skill into your OpenClaw hooks directory:

bash

mkdir -p ~/.openclaw/hooks/cron-scripts-loader
cp <skill-dir>/hooks/cron-scripts-loader/handler.ts \
   ~/.openclaw/hooks/cron-scripts-loader/handler.ts

Then restart the gateway:

bash

openclaw gateway restart

The hook registers itself on gateway.startup and begins watching ~/.openclaw/cron-scripts/ immediately. No further config needed.

2. Create your scripts directory#

bash

mkdir -p ~/.openclaw/cron-scripts

The hook creates this automatically on first startup, but creating it now lets you add scripts before restarting.

Writing scripts#

Every script needs a short frontmatter block in the first 20 lines:

bash

#!/usr/bin/env bash
# name: my-job          # required — shown in gateway logs
# schedule: 0 8 * * *  # required — standard 5-field cron expression
# tz: Europe/Berlin     # optional — IANA timezone (default: UTC)
# timeout: 120          # optional — seconds before SIGTERM (default: 120)

# ... your script here

The cron expression is written in wall-clock time of the given tz — do not convert to UTC yourself.

Then drop it in ~/.openclaw/cron-scripts/ and make it executable:

bash

chmod +x ~/.openclaw/cron-scripts/my-job.sh

No restart needed. The hook watches the directory with fs.watch and hot-reloads within ~300ms of any file change.

Sending notifications from scripts#

Scripts can post to any OpenClaw channel via the local announce API:

bash

# Post to a Telegram user
curl -sf -X POST http://localhost:3000/api/announce \
  -H "Content-Type: application/json" \
  -d "{\"text\": \"Hello from script cron\", \"channel\": \"telegram\", \"to\": \"<YOUR_CHAT_ID>\"}"

Or call external services directly (Telegram Bot API, Slack webhooks, etc.).

Disabling / deleting a job#

• Disable: rename to my-job.sh.disabled — the watcher ignores non- .sh files
• Delete: remove the file — job stops within 300ms, no restart needed

Logs#

All script output is captured and forwarded to the gateway log under [cron-scripts-loader][<name>]:

text

[cron-scripts-loader][my-job] Starting run
[cron-scripts-loader][my-job] some stdout line
[cron-scripts-loader][my-job] Finished ok (exit 0)

Non-zero exit codes are logged as errors but never crash other jobs.

Example script#

See examples/health-check.sh for a minimal HTTP health check that posts failures to Telegram.

Constraints#

• Scripts run as the same user as the gateway process
• Pure bash — no access to the OpenClaw tool system
• If a previous run is still active when the next tick fires, the new run is skipped (croner protect: true) to prevent pile-ups
• Relies on croner from OpenClaw's own node_modules — tested against OpenClaw 2026.x; if croner moves, update the require path in the hook

Context#

This skill was built as a workaround for:

• #80793 — Support script target type in openclaw cron
• #18160 — Direct Exec Mode for Cron Jobs

If those issues are resolved natively, this skill becomes unnecessary. Until then, it's the cleanest way to keep your cron inventory unified in one place without burning model tokens on tasks that are pure bash.