Monitor long-running processes with stability polling and recover-or-stop teardown.
npx skills add https://github.com/btseytlin/ultrapack --skill job-guardian
Please help me install this Agent Skill. Skill: job-guardian Source: https://github.com/btseytlin/ultrapack/tree/main/plugins/up/skills/job-guardian Command: npx skills add https://github.com/btseytlin/ultrapack --skill job-guardian
This Skill prevents long-running jobs from silently failing while you’re away, by enforcing evidence-based liveness and progress checks plus safe, reversible recovery.
Give the AI a launch-and-watch request for a batch job, and it will write a guarding contract, start the job, poll every 270s for progress, and either recover or stop-and-notify with the evidence log.
High-intent search queries and answers about installing and using this skill.
To monitor long-running batch jobs and prevent silent failures, you can use an evidence-based liveness model that polls for progress changes like log growth or throughput movement, ensuring processes run to completion or fail loudly when they stall.
Evidence-based progress detection for remote training runs verifies actual progress changes, such as log growth or step movement, rather than assuming the absence of errors means the job is running fine.
To set up a pre-launch contract for batch processing health checks, you define a written job agreement that confirms real health and progress metrics before committing to long polling intervals.
You can automatically recover crashed pods during remote training by applying predefined recovery playbooks with attempt caps, which execute reversible stop actions before tearing down safely.
The best way to handle failure triage for long-running migrations is to apply a constrained recover-or-stop workflow that uses reversible stop actions, attempt caps, and safe teardown with notifications.
Job monitoring systems often assume a stalled process is still running because they rely on the absence of errors rather than verifying evidence-based progress changes like log growth or throughput movement.
xlelibertySchedule iterative code execution and evaluation tasks for recurring research experiments.
knowletEnforce Design by Contract with pre-conditions, post-conditions, and invariants.
mzqefOrchestrate multi-day reinforcement learning training campaigns with automated monitoring and checkpoint management.
ComposioHQProfiles CPU and memory hotspots in applications and outputs flame graphs and traces.
