Skip to content

Agent Skills for Cron Monitoring

Watchgoose agent skills teach AI coding agents how to audit cron job monitoring, set up checks for scheduled jobs, review monitoring health, and triage missed runs. They work with the Watchgoose MCP server and follow the open Agent Skills format, so the same five skills run in Claude Code, Codex, Cursor, Gemini CLI, Grok Build, and other agents that read skills.

The Five Skills

Skill Use it for
watchgoose-cron-coverage-audit Review the cron jobs, backups, systemd timers, and recurring workers in a repository, and sort each into instrumented, a confirmed gap, or unknown coverage. Works offline with files you provide.
watchgoose-monitor-scheduled-job Draft check settings and a heartbeat wrapper for one scheduled job.
watchgoose-batch-monitor-scheduled-jobs Turn an audit, a list, or a CSV into proposed new checks, then create the ones you approve.
watchgoose-monitoring-health-review Spot paused checks, checks in the new state that are not armed yet (including checks that were just resumed), possible alert routing gaps, and jobs running close to their grace time.
watchgoose-missed-run-triage Summarize what happened when a check goes down or runs late, separating observed facts from possible causes.

Install

The skills are published with the hosted MCP connection at pixeware/watchgoose-plugin.

ChatGPT and Codex: connect the Watchgoose plugin. It includes the five skills.

Gemini CLI:

gemini extensions install https://github.com/pixeware/watchgoose-plugin

Grok Build:

grok plugin install pixeware/watchgoose-plugin --trust

Claude Code, Cursor, and other agents:

npx skills add pixeware/watchgoose-plugin

This command installs only the skills. Connect the hosted server https://mcp.watchgoose.com/mcp separately, as described in the MCP documentation.

Example Requests

  • "Audit ops/crontab and the backup scripts it calls. Which jobs are instrumented, which are confirmed gaps, and which are unknown? Don't change anything."
  • "Draft Watchgoose settings for my daily 02:00 backup with ten minutes of grace time. Show the heartbeat wrapper for review; don't create a check yet."
  • "Is monitoring in this project actually working? Review paused checks, alert routing, and runtime headroom without changing anything."
  • "Turn the audit's confirmed gaps into proposed checks and wait for my approval before creating them."
  • "Explain why my backup check went down last night."

Safety Rules

The skills instruct the agent never to run your scheduled jobs, send test pings, install instrumentation, or read secret stores. They are instructions to the AI agent, not a sandbox, so keep reviewing what the agent proposes. The hard limit is the connection itself: read-only access is the default, and creating, changing, pausing, or deleting checks requires a read-and-write connection that you approve. The skills also tell the agent to ask for your explicit approval before any change, and to create batches one check at a time, stopping at the first error without retrying or rolling back. Keep private ping URLs in your job runner's secret environment, not in chat or source code.

The skills are open source under the MIT license in the watchgoose-mcp repository.