Remote-auto-research

Remote_auto-research

Reusable agent workflows for remote engineering / scientific research.

This repo is designed around one default strategy:

• debug locally
• run heavy / long jobs remotely
• process results remotely or locally as appropriate
• iterate the model/code

It currently provides two main reusable modules:

1. general-code-remote — generic SSH-based remote execution for long-running code
2. pogo-remote — MATLAB -> POGO simulation workflow, post-processing, and iteration

---

Workflow diagrams

Overall remote-research workflow

flowchart TD
    A[Research task arrives] --> B{Task type?}
    B -->|Generic remote code| C[general-code-remote]
    B -->|MATLAB / POGO simulation| D[pogo-remote]

    C --> C1[Edit and debug locally]
    C1 --> C2{Local runtime available?}
    C2 -->|Yes| C3[Run small local validation]
    C2 -->|No| C4[Prepare remote fallback]
    C3 --> C5[Upload via SSH / SCP / rsync]
    C4 --> C5
    C5 --> C6[Run remotely]
    C6 --> C7[Process results locally or remotely]
    C7 --> C8{Validated?}
    C8 -->|No| C1
    C8 -->|Yes| C9[Scale up / batch run]

    D --> D1[Clarify modeling intent]
    D1 --> D2{Which POGO paradigm?}
    D2 -->|Layered composite| D3[Use multilayer composite family]
    D2 -->|Polycrystal / porosity / inclusions| D4[Use paradigm-2 family]
    D2 -->|Complex geometry| D5[Use geometry skeleton family]
    D3 --> D6[Adapt model locally]
    D4 --> D6
    D5 --> D6
    D6 --> D7[Preview / consistency check]
    D7 --> D8[Generate one representative pogo-inp]
    D8 --> D9[Run remotely]
    D9 --> D10[Post-process]
    D10 --> D11{Validated?}
    D11 -->|No| D6
    D11 -->|Yes| D12[Expand to batch study]

POGO automated research loop

flowchart TD
    P1[Research question] --> P2[Classify modeling paradigm]
    P2 --> P3[Choose nearest example family]
    P3 --> P4[Copy example into working directory]
    P4 --> P5[Modify geometry / materials / excitation / defects]
    P5 --> P6[Add preview plots and sanity checks]
    P6 --> P7{Model credible?}
    P7 -->|No| P5
    P7 -->|Yes| P8[Generate one pogo-inp]
    P8 --> P9[Run headless MATLAB locally or remotely]
    P9 --> P10[Run run_pogo.sh remotely]
    P10 --> P11[Collect pogo-hist / pogo-field]
    P11 --> P12[Run post-processing]
    P12 --> P13{Scientific signal usable?}
    P13 -->|No| P5
    P13 -->|Yes| P14{Need more cases?}
    P14 -->|No| P15[Summarize result]
    P14 -->|Yes| P16[Expand parameter sweep / batch run]
    P16 --> P10

---

What this repo is for

This repo is meant to give agents a reusable remote research operating pattern, not just one-off prompts.

Typical use cases:

• running large engineering/scientific scripts on remote Linux servers
• local-debug / remote-run pipelines
• MATLAB model generation followed by remote solver execution
• long simulation campaigns with iterative refinement
• automatic post-processing and batch expansion after one validated case

---

Security principle

This repo must never store:

• private SSH keys
• passwords
• API keys
• your real ~/.ssh/config
• machine-specific private infrastructure details

Agents using these workflows should:

1. first check whether usable SSH configuration already exists locally
2. if not, enter a user-guided SSH setup flow
3. use placeholders/templates inside the repo instead of personal secrets

---

Repository layout

skills/
  remote-auto-research/        umbrella router skill
  general-code-remote/         generic remote execution skill
    references/
      ssh_onboarding.md        SSH setup flow (no secrets stored)
    templates/
      ssh_config.example       Safe SSH config template with placeholders
  pogo-remote/                 POGO workflow skill
    assets/
      run_pogo.sh              Standard remote POGO runner template
    pogoMatlabTools-master/
      visual/export_pogo_field_xdmf.m Headless ParaView XDMF/bin field exporter
    references/                Modeling paradigms, recipes, post-processing
    examples/
      matlab_multilayer_composite/
      matlab_polycrystal_porosity_textured/
      matlab_complex_geometry_skeleton/

adapters/
  claude-code/
    agents/                    Subagent files (.md with YAML frontmatter)
    commands/                  Slash command files (.md with YAML frontmatter)

scripts/
  install.sh                   Installs for codex / openclaw / claude-code

---

POGO modeling coverage

The pogo-remote module currently covers three major modeling paradigms:

1. Layered / sandwich / honeycomb-derived composite models

   • e.g. multilayer composite structures
   • best represented by matlab_multilayer_composite/

2. Polycrystal / porosity / inclusion-rich internal topology

   • e.g. textured grains, pores, inclusions, LoF-type internal features
   • represented by matlab_polycrystal_porosity_textured/
   • includes corrected anisotropic POGO material packing guidance

3. Simple material, complex geometry

   • geometry-first modeling with explicit vertices or parametric geometry
   • represented by matlab_complex_geometry_skeleton/

For large 2D/3D .pogo-field visualization, the POGO workflow includes a headless ParaView export path: run skills/pogo-remote/pogoMatlabTools-master/visual/export_pogo_field_xdmf.m on the remote host to create .xdmf plus raw .bin wavefield files. When a model uses fieldStoreNodes, the exporter writes a sampled point-cloud wavefield instead of invalid full-mesh connectivity.

---

Cross-tool compatibility

This repo is organized so that one source tree can be reused across:

• Codex
• OpenClaw / AgentSkills-compatible tools
• Claude Code

Compatibility summary

Tool	Format used from this repo	Notes
Codex	skills/*/SKILL.md folders	install into ~/.codex/skills/
OpenClaw	skills/*/SKILL.md folders	install into ~/.agents/skills/ or another AgentSkills-compatible path
Claude Code	adapters/claude-code/agents/*.md and adapters/claude-code/commands/*.md	installed as Claude Code subagents and slash commands

---

Install / pull methods

1. Clone the repo

git clone git@github.com:Junfei-TAI/Remote_auto-research.git
cd Remote_auto-research

2. Update later

All three tools can use the same repo update method:

git pull --ff-only

If you installed by symlink, a git pull updates the installed skill automatically.

If you installed by copy, rerun the install command after git pull.

---

Install for Codex

Recommended

./scripts/install.sh codex --mode symlink

This installs the repo skill folders into:

~/.codex/skills/

Copy mode

./scripts/install.sh codex --mode copy

Manual install

Copy or symlink each folder from:

skills/*

to:

~/.codex/skills/

---

Install for OpenClaw

Recommended shared install

./scripts/install.sh openclaw --mode symlink

This installs into:

~/.agents/skills/

If your OpenClaw setup prefers another AgentSkills-compatible directory, use:

./scripts/install.sh openclaw --mode symlink --target-dir /your/preferred/skills/path

Copy mode

./scripts/install.sh openclaw --mode copy

Alternative workspace usage

If your OpenClaw setup loads workspace skills directly, you can also run from this repo and let it read the local:

skills/

folder.

---

Install for Claude Code

Claude Code does not use the same SKILL.md folder format directly. This repo provides adapted Claude Code files in:

adapters/claude-code/agents/
adapters/claude-code/commands/

Recommended

./scripts/install.sh claude-code --mode symlink

This installs:

~/.claude/agents/
~/.claude/commands/

Copy mode

./scripts/install.sh claude-code --mode copy

Manual install

Copy or symlink:

adapters/claude-code/agents/*.md
adapters/claude-code/commands/*.md

to:

~/.claude/agents/
~/.claude/commands/

---

Quick usage ideas

General remote code task

Ask the agent to:

• patch locally
• validate locally if possible
• upload to server
• run remotely
• fetch or remotely post-process results

POGO task

Ask the agent to:

• clarify the modeling intent
• choose the correct paradigm
• adapt a bundled example family
• run one representative case first
• generate .pogo-inp
• run run_pogo.sh remotely
• process .pogo-hist / .pogo-field
• decide whether to iterate or scale up

---

Important notes for POGO users

• Example MATLAB/POGO code in this repo is intended as a sanitized starting point, not a fixed universal model.
• For anisotropic material packing, use the corrected POGO packing pattern in:

skills/pogo-remote/examples/matlab_polycrystal_porosity_textured/material_packing_reference.m

• Keep large generated artifacts such as:
  • .pogo-inp
  • .pogo-block
  • .pogo-hist
  • .pogo-field
  • figures / temporary outputs

out of git.

---

Current status of the example families

matlab_multilayer_composite/

• strongest validated family at the moment
• suitable for layered composite / sandwich-style studies

matlab_polycrystal_porosity_textured/

• cleaned and corrected as a paradigm-2 reference family
• anisotropic packing corrected
• some original external research dependencies may still need to be supplied depending on the exact workflow

matlab_complex_geometry_skeleton/

• geometry-first starter skeleton
• useful when vertices, surfaces, or meshing strategy are the main challenge

---

Recommended maintenance workflow

After changing the repo:

git pull --ff-only
./scripts/install.sh codex --mode symlink
# or openclaw / claude-code if you installed by copy

If using --mode symlink, usually only:

git pull --ff-only

is needed.