Retire an automation nobody uses
Automations outlive their purpose and keep costing upkeep and attention. Retire them deliberately: confirm nobody uses the output, remove the code and schedule, and leave a record.
Procedure
- Find candidates mechanically, not by opinion: jobs with no consumer, no alert actioned in 90 days, or an owner that has changed three times.
- Confirm the output has no downstream: search code, dashboards, and other jobs for the artifact it produces (the table, file, message, or endpoint).
grep -r "nightly_recon_summary" --include='.sql' --include='.py' .
- Check the alert history: an alert fired for 90 days and never actioned is evidence the job's signal is unwanted.
- Ask the recorded owner once, with a deadline: "Retiring X on the 15th unless you object" — a silent owner is consent, but the notice must exist.
- Quiesce before deleting: pause the job for one full cadence and watch for the complaints that prove it was used.
- Remove in order: disable the schedule and alert first (reversible), then delete the code and config after the quiet period.
- Clean up what it created: its status row, its runbook, its resources (containers, output buckets, cron entries), so no orphans remain.
- Leave a tombstone: a one-line record of what it was, why removed, and the date, in the changelog or a
retired.tsv.
- Revoke its credentials and any allowlisting tied to it, so a retired job's keys do not linger.
- If a consumer surfaces after retirement, bring the job back from version control rather than from memory.
Pitfalls
- Deleting on suspicion without a search for consumers or a notice period.
- Retiring the job but leaving its credentials, alert, and status row alive.
- A quiet period too short to catch a monthly consumer.
- Removing a job whose output feeds a compliance or audit requirement.
- No tombstone, so a future engineer re-adds the same job in six months.
- Turning off the alert only, so the job keeps running invisibly and burning cost.
Verification
# no references to the retired artifact remain
grep -rn "nightly_recon_summary" --include='*.sql' --include='*.py' . | wc -l # 0
crontab -l | grep -c nightly-reconcile # 0
# pass: no consumers, no schedule, credentials revoked, tombstone written
Report the job, the consumers or their absence, the notice sent and its deadline, what was removed, and the tombstone line.
1---2name: retire-an-automation-nobody-uses3description: Use when an automation may no longer be needed because its output is unread, its alert ignored, or its task obsolete. Confirms, then removes it cleanly.4metadata:5 source: https://skillmd.com/skills/jiayaoqijia/retire-an-automation-nobody-uses6---78# Retire an automation nobody uses910Automations outlive their purpose and keep costing upkeep and attention. Retire them deliberately: confirm nobody uses the output, remove the code and schedule, and leave a record.1112## Procedure13141. Find candidates mechanically, not by opinion: jobs with no consumer, no alert actioned in 90 days, or an owner that has changed three times.152. Confirm the output has no downstream: search code, dashboards, and other jobs for the artifact it produces (the table, file, message, or endpoint).16 grep -r "nightly_recon_summary" --include='*.sql' --include='*.py' .173. Check the alert history: an alert fired for 90 days and never actioned is evidence the job's signal is unwanted.184. Ask the recorded owner once, with a deadline: "Retiring X on the 15th unless you object" — a silent owner is consent, but the notice must exist.195. Quiesce before deleting: pause the job for one full cadence and watch for the complaints that prove it was used.206. Remove in order: disable the schedule and alert first (reversible), then delete the code and config after the quiet period.217. Clean up what it created: its status row, its runbook, its resources (containers, output buckets, cron entries), so no orphans remain.228. Leave a tombstone: a one-line record of what it was, why removed, and the date, in the changelog or a `retired.tsv`.239. Revoke its credentials and any allowlisting tied to it, so a retired job's keys do not linger.2410. If a consumer surfaces after retirement, bring the job back from version control rather than from memory.2526## Pitfalls2728- Deleting on suspicion without a search for consumers or a notice period.29- Retiring the job but leaving its credentials, alert, and status row alive.30- A quiet period too short to catch a monthly consumer.31- Removing a job whose output feeds a compliance or audit requirement.32- No tombstone, so a future engineer re-adds the same job in six months.33- Turning off the alert only, so the job keeps running invisibly and burning cost.3435## Verification3637 # no references to the retired artifact remain38 grep -rn "nightly_recon_summary" --include='*.sql' --include='*.py' . | wc -l # 039 crontab -l | grep -c nightly-reconcile # 040 # pass: no consumers, no schedule, credentials revoked, tombstone written4142Report the job, the consumers or their absence, the notice sent and its deadline, what was removed, and the tombstone line.