Documentation
  1. 01Getting started
  2. 02Next.js and Vercel
  3. 03Servers and scripts
  4. 04Schedules, grace and timeouts
  5. 05What it catches
  6. 06Alerts
  7. 07Stores
  8. 08Dashboard and API
  9. 09MCP server
  10. 10Agent skill
  11. 11AI triage
  12. 12API reference
  13. 13Limits and design notes

Agent skill

Agent skill

The repository ships skills/cronwatch/SKILL.md, a skill for Claude Code (and other agents that read the same format). With it installed, asking for "monitor this cron job" or "why did the nightly job fail" gets a workflow instead of a guess.

Install

Copy the folder into a project or your user skills:

# in a project
mkdir -p .claude/skills
curl -sL https://raw.githubusercontent.com/phillips-jon/cronwatch/main/skills/cronwatch/SKILL.md \
  -o .claude/skills/cronwatch/SKILL.md --create-dirs

# or for every project
curl -sL https://raw.githubusercontent.com/phillips-jon/cronwatch/main/skills/cronwatch/SKILL.md \
  -o ~/.claude/skills/cronwatch/SKILL.md --create-dirs

What it teaches

Adding monitoring. Find where the job runs, pick the store the app already has, declare the job once with its real schedule and timezone, wrap the work, mount the routes, make sure something calls the check, add a channel. It tells the agent to ask before adding dependencies and to keep job names stable.

Investigating. With the MCP server configured: list jobs, read the failing run’s error and output before changing code, fix the cause rather than the monitor, silence only during a known fix, and confirm recovery after deploying.

The vocabulary. What missed, failed, stuck, slow, over budget and recovered mean, so the agent explains an alert correctly.

Pair it with the MCP server and the agent can go from "the digest didn’t arrive" to a pull request with the fix and a confirmation that the next run recovered.