메뉴
Run blueprints on the Magnus cloud platform via the `magnus` CLI. Use this skill when you need to execute remote computations on GPU/CPU clusters.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Run MadGraph5_aMC@NLO event generation for particle physics simulations via Magnus cloud. Triggers when the user wants to generate Monte Carlo events, simulate collider processes, or run MadGraph5. Supports Pythia8 parton shower and Delphes detector simulation.
Compute dark matter observables (relic density, direct detection, indirect detection) via micrOmegas on the Magnus cloud. Triggers when the user wants to calculate Ωh², DM-nucleon cross sections (SI/SD), annihilation ⟨σv⟩ into SM final states, or indirect-detection γ/ν/e⁺/e⁻ spectra. Consumes a CalcHEP model (from the calchep-generator skill) plus a user-supplied main.c.
Generate CalcHEP model directories from FeynRules .fr files via Magnus cloud. Triggers when the user has a validated .fr model and needs a CalcHEP model for micrOmegas or CalcHEP.
Generate a self-contained reproduction guide from a completed analysis pipeline run. Triggers when the user asks to "create a reproduction guide", "write a reproduction package", "make this reproducible", or "create scripts for reproducing this analysis".
Generate FeynRules .fr model files from LaTeX Lagrangian descriptions. Triggers when the user provides a Lagrangian in LaTeX notation and wants it converted to a FeynRules model file for particle physics simulations.
Validate FeynRules models for correctness. Two capabilities: (1) Mathematica-based physical consistency checks on .fr files (Hermiticity, diagonal mass/quadratic terms, kinetic term normalization), (2) MadGraph5 import test on generated UFO directories to catch Python syntax errors and structural issues.
| name | magnus |
| description | Run blueprints on the Magnus cloud platform via the `magnus` CLI. Use this skill when you need to execute remote computations on GPU/CPU clusters. |
Magnus executes blueprints on remote GPU/CPU clusters. The CLI is self-documenting — use --help at each level to discover what's available:
magnus --help # top-level commands and shortcuts
magnus run --help # CLI options for blueprint execution
magnus job --help # job inspection and management
magnus blueprint schema <blueprint-id> # parameter schema for a specific blueprint
pip install "magnus-sdk>=0.6.0"
magnus config # check current connection — skip setup if already configured
If not configured, run magnus login (saves to ~/.magnus/config.json, all shells pick up immediately):
magnus login # interactive
magnus login <site> -a <address> -t <token> # non-interactive (for scripts and agents)
| Priority | Source | Use case |
|---|---|---|
| 1 (highest) | Environment variables MAGNUS_ADDRESS, MAGNUS_TOKEN | CI/CD pipelines, automation scripts |
| 2 | ~/.magnus/config.json (via magnus login) | Recommended for interactive use |
| 3 (lowest) | Built-in default site | Fallback |
Do not use environment variables for interactive work. Inline MAGNUS_ADDRESS=... magnus run ... is non-portable and not persisted. If magnus config shows an env-override warning, remove the export lines from your shell profile (.bashrc, .zshrc, or system environment variables on Windows) to let magnus login take effect.
Run magnus config first to verify your connection. If not configured, use magnus login once — credentials persist across all shells and subsequent commands. Do not inline MAGNUS_ADDRESS/MAGNUS_TOKEN environment variables before each command.
magnus list # discover available blueprints
magnus blueprint schema <id> # inspect parameters, types, defaults
magnus run <id> [--] --key value ... # execute and wait for result
-- separatorDivides CLI options (left) from blueprint arguments (right). Optional — without it, all arguments route to the blueprint and CLI options use defaults. See magnus run --help for available CLI options.
When a blueprint parameter is typed FileSecret, passing a local file path triggers automatic upload. Files and directories both work. Check magnus blueprint schema <id> to see which parameters accept file paths.
Blueprints that produce output files write a magnus receive command to MAGNUS_ACTION. With --execute-action true (default), magnus run auto-executes this receive command.
magnus job result <job-id> # structured result (same as magnus run output)
magnus logs <job-id> # stdout + stderr from cloud execution
magnus status <job-id> # job status
magnus kill <job-id> # terminate a running job
magnus jobs # list recent jobs with IDs
<job-id> is the absolute job ID printed by magnus run (e.g. abc123). Negative indices also work as a shorthand (-1 = most recent, -2 = second most recent), but they are scoped to your account, not your session — if another agent or session submits a job under the same account, the indices shift. In multi-step pipelines, prefer capturing the job ID from magnus run output and passing it explicitly.
Blueprint results typically contain at minimum success (bool) and message (str).
If magnus run returns a non-success result:
magnus job result <job-id> — structured error detailsmagnus logs <job-id> — full execution outputJobs continue server-side even if your client disconnects — interrupting magnus run (Ctrl-C) or a network drop only detaches the local client, the job keeps running. Do not re-submit — instead, reconnect:
magnus status <job-id> (or sleep 30 && magnus status <job-id> to wait out a transient outage)magnus job result <job-id> to read MAGNUS_RESULTmagnus receive command from MAGNUS_ACTION to download outputsUse magnus kill <job-id> to terminate a job that was submitted with wrong parameters to free up cluster resources.