Post-failure autopsy
When a run dies, Smithers does not just leave a corpse. It runs an autopsy on the spot, works out what killed it, and hands you a fix waiting behind an approval gate.
Why this exists
A transient network blip, a rate limit, one bad edit, and the overnight job is dead. You wake up to a stalled run, a burned quota window, and a morning spent reading logs to reconstruct what went wrong.
How it works
The moment a run fails, Smithers launches a post-failure autopsy workflow on its own. No cron, no babysitting.
The autopsy inspects the run with real tools: what step, what error, what state. It writes a diagnosis, not a guess.
A fix is proposed behind an approval gate. Nothing changes your code until you say so. Approve, and it retries.
What you get
A failed run launches the autopsy with no wiring. You wake up to a diagnosis, not a mystery.
The autopsy inspects the actual run and names the failure, so the fix targets the real problem.
Remediation is a proposal behind an approval gate. Your codebase is safe while you sleep.
A medic step patrols your monitored runs and applies the same treatment across everything at once.
Let the autopsy do the forensics and hand you the fix. Smoke-test it now with the fail-probe.