Thin global entry point for enabling the full ml-research-skills collection only where it belongs.
Do not solve ML research tasks directly. Decide whether the current workspace should receive the full project-local skill bundle, then route to the right setup action.
Distribution Rule
Use a two-layer install model:
Global agent runtime:
ml-research-bootstrap only
Initialized ML research project:
full ml-research-skills bundle installed locally
Generic non-ML project: no ml-research-skills bundle
This keeps ordinary software projects free of ML-research routing while preserving the full system inside real research projects.
## Workflow
1. Detect scope:
```bash
git rev-parse --show-toplevel
git rev-parse --git-common-dir
- Classify the request:
- New ML research project: route to
project-init. - Existing ML research project: install or refresh the full bundle from the project root.
- This skill repository: maintainer work may use the full bundle for validation and runtime checks.
- Generic non-ML project: do not install the full bundle; explain that the collection is project-local for ML research.
- For an existing ML research project, prefer:
cd <ML_PROJECT_ROOT>
npx skills add a-green-hand-jack/ml-research-skills
-
For a new ML research project, use
project-initto create the control root, memory, component repos, and project-local agent guidance before enabling the full bundle. -
After the full bundle is local to the project, ML research workflow selection starts at
ml-research-routeronly inside that project.
Global Install Command
Recommended default:
npx skills add a-green-hand-jack/ml-research-skills -g -a codex claude-code -s ml-research-bootstrap -y
If the agent must create new project roots from a global runtime and the skills CLI cannot resolve paired skills automatically, also install project-init globally:
npx skills add a-green-hand-jack/ml-research-skills -g -a codex claude-code -s project-init -y
Reserve full global installs for maintaining this skill repository or debugging installed-runtime behavior.
Boundaries
- Do not route ordinary codebase tasks into
ml-research-router. - Do not install the full bundle into a repo just because it contains Python, LaTeX, or data files.
- Do not perform paper writing, experiment launch, citation audit, or remote execution from this skill; enable the project-local bundle first, then route through the appropriate project-local skill.