Resources · RCC ClusterDocs

RCC Guardians and AI-assisted support

RCC uses automation to reduce repetitive support and operations work while keeping consequential actions bounded.

Getting help today

Contact the RCC team in the IKIM Cluster channel on Mattermost. Do not post credentials, private keys, restricted sample information, or other sensitive project data there.

Coding agent

Service status — not yet released: the RCC help assistant and coding agent are not yet available to users. This section describes the intended service.

When released, the RCC coding agent is intended to explain RCC concepts, find documentation, interpret common Slurm states/errors, diagnose supported environment/container problems, explain transfer/storage guidance, prepare project requests, and escalate with useful evidence.

It can also help convert sanitized shell scripts or command collections into a dry-run-capable Snakemake or Nextflow workflow with explicit inputs, outputs, resources, tests, and a pinned Apptainer runtime. Start with the safe conversion handoff. Scientific method, parameters, reference data, and interpretation remain the project team's responsibility.

Guardians

Guardians observe scheduler, service, hardware, storage, and policy state. Observation and action remain separate.

Constrained actors

A Guardian does not receive unrestricted root simply because it can diagnose a problem. Supported remediation uses named operations such as:

drain node with reason
restart approved service
prepare archive plan

rather than arbitrary shell access.

Escalation is valid

Success includes safe resolution, a useful explanation, preparation of a guarded request, or correct recognition that a human operator is required.

For project actions, Guardians use the same authority model as users and delegated proxies.

Never paste protected research data, credentials, private keys, or MFA secrets into a support conversation unless an explicitly approved process requires it.