Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add PV-Bhat/dtu-hpc-mcp --skill dtu-hpcgit clone --depth 1 https://github.com/PV-Bhat/dtu-hpc-mcpWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/pv-bhat/dtu-hpc-mcp/dtu-hpc)<a href="https://agentmods.dev/skills/pv-bhat/dtu-hpc-mcp/dtu-hpc"><img src="https://agentmods.dev/badge/skills/pv-bhat/dtu-hpc-mcp/dtu-hpc/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/pv-bhat/dtu-hpc-mcp/dtu-hpc"><img src="https://agentmods.dev/badge/skills/pv-bhat/dtu-hpc-mcp/dtu-hpc.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00088 | $0.02388 |
| Opus 5 | $0.00044 | $0.01194 |
| Sonnet 5 | $0.00018 | $0.00478 |
| Haiku 4.5 | $0.00009 | $0.00239 |
Grade A, and why
dtu-hpc scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 174 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Operating DTU HPC through dtu-hpc
You are operating a real, shared university cluster through a constrained control plane. The human you work for is the research director: they set goals, budgets, and constraints; you handle execution. The account you act under is theirs, and they are personally accountable for everything you do — act like a careful employee, not a power user.
Mental model
You do not have a shell. You have named operations (a vending machine, not a
terminal), served by a local daemon (dtuhpcd) that holds one authenticated
SSH session. Three consequences:
- If
hpc_statusreports the daemon is down, stop. Only the human can start and authenticate it (dtu-hpc up). Tell them; do not retry in a loop. - Everything you do is written to a local audit log the human reads. Denials are logged too. Never try to rephrase a request to slip past a denial — a denial means the policy layer is working; report it instead.
- The session may drop (VPN, idle). The daemon reconnects automatically on the next call; one transient failure is not an outage.
The one constraint that shapes everything: 24h GPU walltime
Every GPU queue kills jobs at 24 hours. Any training run longer than that MUST be submitted as a chain: N dependent jobs, each resuming from checkpoints.
- Segments are linked with
ended(<prev>), so a crashed segment still lets the next one start from the last checkpoint. This means a chain can quietly "succeed" while making no progress — always verify between segments that the checkpoint actually advanced before letting a chain continue. - Jobs receive SIGUSR2 ten minutes before the walltime kill. The training
script must checkpoint on SIGUSR2 and support
--resume-from. If the code you are launching lacks these, fix that FIRST — a chain without them loses work at every boundary. - Plan checkpoint frequency around segment boundaries, not convenience.
Queue selection
| Need | Queue | Notes |
|---|---|---|
| Large model training | gpua100 |
2× A100 PCIe per node. Mixed fleet: 4 nodes 40GB + 6 nodes 80GB, and there is NO selector to force 80GB. If the run needs >40GB per GPU, check hpc_gpu_models("gpua100") first and fail fast in-job on a 40GB card — or use gpul40s for a deterministic 48GB |
| Deterministic 48GB | gpul40s / gpua40 |
2× L40S / 2× A40-NVLink |
| Small/debug GPU runs | gpuv100 |
2× or 4× V100, 16/32GB. select[gpu32gb] and select[sxm2] (NVLink) are the ONLY documented hardware selectors, both V100-only |
| CPU work, preprocessing | hpc |
72h cap; never run CPU-heavy preprocessing on a GPU queue |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 174 lines · 88 tokens per session scan A 755f5e20914a
dtu-hpc is a skill published in the GitHub repository PV-Bhat/dtu-hpc-mcp (0 stars, last pushed 1mo ago), licensed MIT. It adds 88 tokens to every session and 2,388 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
using-drac-clusters
Use when working on a Digital Research Alliance of Canada (DRAC, Alliance, formerly Compute Canada) HPC cluster such as Narval, Rorqual, Fir, TamIA, Killarney, Vulcan, Nibi or Trillium — including ssh to .alliancecan.ca failing with "Permission denied (keyboard-interactive)", sinfo/sacctmgr reporting "command not…
latchbio-integration
Build, register, debug, and operate bioinformatics workflows on Latch using the Python SDK, CLI, Latch Data and Registry, Nextflow, Snakemake, programmatic execution, and Latch MCP. Use when authoring or deploying Latch workflows, configuring resources or interfaces, moving data, integrating Registry, or launching and…
remote-compute-ssh
Evaluate and use SSH Remote Compute before choosing where to run GPU, high-memory, parallel, batch, model-inference, bioinformatics, or other long-running scientific work; supports short remote commands and asynchronous jobs with automatic harvest and analysis.
modal-compute
Run explicitly chosen research benchmark or replication jobs on Modal's serverless infrastructure. Use when a Feynman research workflow needs burst remote GPU compute and the Modal CLI is available.
dnanexus-integration
DNAnexus cloud genomics platform. Build apps/applets, manage data (upload/download), dxpy Python SDK, run workflows, FASTQ/BAM/VCF, for genomics pipeline development and execution.
latchbio-integration
Latch platform for bioinformatics workflows. Build pipelines with Latch SDK, @workflow/@task decorators, deploy serverless workflows, LatchFile/LatchDir, Nextflow/Snakemake integration.