Flake Hunter

The homepage Trigger it card labeled Cron is this example. A schedule is optional, not the product. The product is a package you own that can run while your laptop is closed, with no model in the loop and no tokens spent.

Live public package: @kentcdodds/e2e-flake-hunter. Fork that listing when you want the same hunt on your repo. The primitive is documented in Jobs, workflows, and webhooks.

The job

Every day at 4am America/Denver, the published package scans the last 24 hours of GitHub Actions jobs on kentcdodds/kody. If any job indicates e2e flakiness, it starts one investigation. A clean day is silent.

That is the factory loop from How Kody works aimed at CI noise: ask once, save the scan, trigger it on a cron.

Kent's copy also spawns one Cursor Cloud Agent when a flake signal is present. Obvious fixes ship through the repo's ship-pr path. A non-obvious flake becomes a written recommendation emailed to Kent with a link to the investigating agent. Your fork can stop at "scan and notify" if you do not want an agent spawned.

Package shape

Declare the schedule in package.json#kody.jobs so it travels with the code it runs:

{
	"kody": {
		"jobs": {
			"daily": {
				"entry": "./src/daily.ts",
				"schedule": { "type": "cron", "expression": "0 4 * * *" },
				"timezone": "America/Denver",
				"enabled": false
			}
		}
	}
}

Publish with "enabled": false. The entry is a no-argument wrapper. The usual shape is a thin module that calls a callable export and notifies only when there is news:

import sweep from './sweep.ts'

export default async function daily() {
	return await sweep()
}

Keep the scan itself callable from execute so you can prove it before the clock does:

import scan from 'kody:@you/e2e-flake-hunter/scan'

export default async function main() {
	return await scan()
}

• Shared implementation accepts an optional cursor (since / last seen run id) and returns { flakes, message }.
• A read-only ./scan export never spawns work and never sends mail. Use it from chat and from the scheduled wrapper's dry path.
• A ./sweep export runs the scan, records the newest run id in packageStorage(), and only then starts one investigation or emailSend.
• emailSend only mails the account's own verified address. Skip it when flakes.length === 0. An empty digest every morning trains people to ignore the real one.
• Each job run gets a job-scoped scratch bucket. The cursor belongs in packageStorage(), not in the scratch bucket.

jobRunNow fires one run for debugging. jobUpdate flips enabled after that run looks right. Name and source stay in the repo.

Example prompts

Paste one of these into an agent already connected to your Kody account.

Build one

Search Kody for flake hunter and official job guides. I want a daily cron that scans the last 24 hours of GitHub Actions jobs on this repo for e2e flakes. Stay quiet when the day is clean. If a flake signal is present, record it and mail me once. Publish with the job disabled, invoke the wrapper from execute, then enable it only after that run looks right.

Fork the public one

Open https://kody.codes/@kentcdodds/e2e-flake-hunter, fork it into my account, and adapt it to my repo. Keep the daily job disabled until we have invoked ./daily once. Prefer ./scan for the first check. Do not disable a live job on Kent's package.

Operate an existing copy

Import kody:@me/e2e-flake-hunter/status and tell me whether today's sweep ran, whether the kill switch is on, and what the last outcome was. If I want it quiet for a bit, use ./pause rather than disabling the job.

What you see

The homepage card is a tile with the kicker Cron and the title Flake Hunter. It links here so the marketing example and the docs stay the same story.

Where to go next

• Jobs, workflows, and webhooks — when to pick a job versus a webhook, a subscription, or the inbox.
• Package lifecycle — test a scheduled wrapper before enabling it.
• How Kody works — the quieter daily-email loop this example is built on.
• Sentry Issues — the webhook sibling on the same homepage row.