What problem does it solve? Long-running Slurm benchmark jobs for the ECODA scRNA-seq pipeline can be lost or left in ambiguous states when SSH sessions drop, and ad-hoc retries risk duplicate launches, corrupted artifacts, or unverified results. This Skill enforces a durable, fail-closed gate lifecycle so every ECODA compute run on Yggdrasil is prepared, launched, monitored, audited, and reviewed exactly once with full provenance. ## Core Features & Use Cases - Durable gate lifecycle: Drives the global durable-hpc-gate CLI through reconcile, prepare, launch, wait, inspect, and status subcommands with atomic manifests and event-driven waiting instead of polling. - Repository and host invariants: Validates the Yggdrasil checkout, canonical scratch/NAS roots, slurm_config.sh settings, immutable source snapshots, and run-bound runtime identities before any launch. - Serialization and lineage control: Uses explicit serialization groups as cross-manifest mutexes and reviewed dependency manifests so benchmark waves run safely in parallel or sequence. - Terminal audit and review: Performs a single strict sacct accounting query, run-scoped ecoda_run_audit.sh validation, checksum verification, and a mandatory Luna Max reviewer approval before release. - Use Case: Launch a full-cohort ECODA benchmark wave on Yggdrasil, disconnect your laptop, and later resume to a verified, reviewer-approved COMPLETED gate with audited scheduler IDs and artifact checksums. ## Quick Start Ask the agent to prepare and launch a durable ECODA gate on Yggdrasil for a specific pipeline stage using the checked-in profile and an explicit gate ID, then wait and inspect it to completion.
