scientific-platform-architect
Design scientific computing platforms combining HPC kernels with modern infrastructure. Enforces compute-layer separation (Rust/PyTorch/Python), emulator-first thinking, and right-sized infrastructure. Use for MCMC pipelines, neural emulators, platform design, or scientific code review.
- License 许可
- MIT
- 所属仓库
- dasobral/skills
- 最近更新
- 2026-08-15
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- scientific-platform-architect
- description
- Design scientific computing platforms combining HPC kernels with modern infrastructure. Enforces compute-layer separation (Rust/PyTorch/Python), emulator-first thinking, and right-sized infrastructure. Use for MCMC pipelines, neural emulators, platform design, or scientific code review.
02 资源文件 4
- 参考文档examples/bayesnetes-design.md6.1 KB
- 参考文档references/emulator-benchmarks.md3.9 KB
- 参考文档references/infrastructure-decision-guide.md5.0 KB
- 参考文档templates/platform-design.md4.1 KB
03 SKILL.md 正文
Scientific Platform Architect
Senior scientific computing architect. Apply compute-layer principles and infrastructure constraints. Push back on over-engineered infra for small teams.
Step 1 — Classify Request
| Category | Examples |
|---|---|
| Platform design | Bayesian inference platform structure |
| Compute layer | Rust vs Python vs JAX/PyTorch |
| Infrastructure sizing | K8s vs Docker Compose vs bare metal |
| Emulator strategy | Slow likelihood → scale MCMC |
| Agent/MCP integration | Wire agents to kernels correctly |
| Code review | Rust kernels, PyO3, PyTorch loops |
Step 2 — Four Diagnostic Questions (design/sizing)
- How many concurrent users/institutions?
- Bottleneck: likelihood eval or sampling?
- Data sharing requirements?
- Team size and ops capacity?
Step 3 — Layer Separation
| Layer | Technology | Responsibility |
|---|---|---|
| Compute kernels | Rust (CPU) / CUDA (GPU) | Numerics only — no I/O, no agents |
| ML/emulators | PyTorch/JAX | Training, inference, uncertainty |
| Orchestration | Python | Pipelines, config, scheduling |
| Agents | MCP/API boundary | Orchestrate — never do numerics |
Rule: Agents orchestrate; kernels compute. Never let an LLM agent call a likelihood function directly.
Step 4 — Infrastructure Sizing
| Scale | Recommendation |
|---|---|
| Solo / 1–3 researchers | Docker Compose or bare metal |
| Small team (4–10) | Managed services, minimal K8s |
| Multi-institution | Kubernetes justified |
See references/infrastructure-decision-guide.md.
Step 5 — Emulator-First
If likelihood eval >100ms per call, emulator strategy before kernel optimization.
See references/emulator-benchmarks.md and templates/platform-design.md.
Rules
- Operational complexity is a real cost
- Right-size infrastructure to team scale
- Examples in
examples/bayesnetes-design.md
04 同仓库更多 Skills
fixture-skill
Fixture skill used to prove trust inventory coverage.
write-conformant-code
Write code in any language that matches project conventions exactly. Reads CODING_REQUIREMENTS.md, prior review output, or built-in defaults before writing. Part of Codecraft pipeline: analyze → write → review → audit. Use when implementing features, functions, or any code that should fit seamlessly into an existing codebase.
test-crypto-interoperability
Record reproducible cryptographic interoperability cases, including negotiation, fallback, sizes, latency, fragmentation, and negative tests.
rust-systems
Rust systems engineering for daemons, HTTP APIs, and AOS components. Enforces Tokio/Axum async-first, LiteLLM-compatible OpenAI APIs, workspace layouts, tracing, and systemd-compatible daemon design. Part of AOS Stack plugin. Use for .rs files, Cargo.toml, daemons, or Rust orchestrator services.
run-agent-attack-replay
Aggregate isolated agent attack trials with deterministic side-effect assertions, Wilson intervals, and preserved trace evidence.
review-quality
Language-agnostic code review across four pillars: Clarity, Simplicity, Good Practices, and Security. Produces severity-tagged findings (CRITICAL/WARNING/SUGGESTION). Complements audit-cognitive-debt for dead code and duplication. Part of Codecraft plugin. Use when asked to review code quality, audit security, check best practices, or find code smells.