eegprep-extension-development - SKILL.md Agent Skill

name: eegprep-extension-development description: Build EEGPrep external extension packages with SDK registration, declarative menus, pop_* console/history behavior, packaged help/data, distribution choices, tests, GUI visual parity, and catalog-readiness. Use when Codex or a researcher needs to create, port, review, package, or submit an EEGPrep extension outside the core package.

EEGPrep Extension Development

Use this skill to build extension packages that work as standalone Python packages and feel natural inside EEGPrep, including for researchers migrating from EEGLAB.

Start Here

Read AGENTS.md and follow .agents/skills/eegprep-feature-development/SKILL.md for repo workflow, notes, tests, and PR expectations.
Inspect src/eegprep/extensions.py and docs/source/api/extensions.rst before writing registration code. Use real SDK symbols, especially ExtensionSpec, ExtensionAction, ExtensionMenu, ExtensionPopFunction, ExtensionResource, ExtensionDependency, LazyImport, validate_extension_spec, and the eegprep.extensions entry-point group.
If porting an EEGLAB plugin or workflow, inspect the matching EEGLAB MATLAB source during development. Do not make runtime extension code depend on a local EEGLAB checkout.
Check examples/extensions/ for templates, scaffolding, and focused example packages. Reference or use them where available; do not duplicate their full implementation inside custom extension code.

Plan the Extension

Define:

Distribution path: local/private editable, GitHub-only, PyPI, or private package index.
Trust posture: experimental, private lab use, public uncurated, or catalog submission. Curated means catalog-reviewed and mechanically validated, not scientific endorsement.
Contributions: actions, menus, pop_* functions, help resources, package data, console aliases, optional dependencies, and large model files.
EEGPrep version contract through eegprep_requires and extension API version.
Tests that prove discovery, validation, lazy imports, resources, history, console behavior, and GUI behavior where applicable.

Prefer the smallest SDK surface that expresses the feature. If the extension needs to mutate EEGPrep internals directly, stop and add or request an SDK surface instead.

Register Through the SDK

Create a normal Python package with pyproject.toml:

[project]
name = "eegprep-ext-foo"
version = "0.1.0"
dependencies = ["eegprep"]

[project.entry-points."eegprep.extensions"]
foo = "eegprep_ext_foo.register:register"

Keep register() lightweight:

from eegprep import ExtensionAction, ExtensionMenu, ExtensionPopFunction, ExtensionSpec, LazyImport


def register():
    return ExtensionSpec(
        name="foo",
        display_name="Foo",
        version="0.1.0",
        package_name="eegprep-ext-foo",
        eegprep_requires=">=0.2",
        actions=(
            ExtensionAction("foo.run", LazyImport("eegprep_ext_foo.actions", "run")),
        ),
        menus=(ExtensionMenu(("Tools", "Foo"), "foo.run", "Run Foo"),),
        pop_functions=(
            ExtensionPopFunction("pop_foo", LazyImport("eegprep_ext_foo.pop_foo", "pop_foo")),
        ),
    )

Use LazyImport for processing modules, dialogs, models, and optional dependencies. A registration function may import the SDK and small metadata; it must not import heavy code, start Qt, read data files, or call EEGLAB.

Menus, Help, Console, and History

Declare menus with ExtensionMenu; do not mutate Qt menu objects or patch EEGPrep menu builders.
Keep user-facing actions compatible with EEGPrepSession. GUI actions should store datasets, add history, and notify changes through session helpers.
Implement user-facing pop_* functions with return_com=True; return (EEG, com) when history should be recorded.
Preserve EEGLAB-facing command strings and indexing semantics while using Python indexing internally.
Package GUI Help and pophelp text as Markdown resources and declare them with ExtensionResource.
Update the relevant Sphinx docs when an extension changes user-facing menus, workflows, console/history behavior, optional dependency expectations, package data, curation metadata, or troubleshooting guidance. Keep the docs centered on standalone EEGPrep behavior, with EEGLAB comparisons only where they clarify migration or compatibility.
Put data/model resources in the package and access them with importlib.resources or declared ExtensionResource records. Do not read from src/eegprep/eeglab at runtime.

Dependencies and Large Files

Declare required Python packages normally in pyproject.toml.
Use ExtensionDependency(package, version_spec, optional=True) only for optional behavior. Required missing dependencies should produce missing_dependency, not a late crash.
For large models, choose an explicit path: package extra, private index, Git LFS-backed package source, or an extension-owned first-run download. Document the choice and never imply EEGPrep hosts model artifacts.

GUI Extensions

For dialogs, windows, or menu UX:

Use .agents/skills/eeglab-gui-visual-parity/SKILL.md to compare labels, alignment, order, defaults, enabled states, buttons, and workflow against EEGLAB where an EEGLAB reference exists.
Exercise mixed GUI plus eegprep-console flows: GUI action, inspect EEG, ALLEEG, CURRENTSET, LASTCOM, and ALLCOM; then run the pop_* function from the console and verify GUI/history refresh.
Use .agents/skills/gui-agent-flow-qa/SKILL.md only when explicitly asked for full GUI agent QA or when the project workflow requires it.

Tests and Checks

Write focused tests for:

Entry-point discovery through ExtensionRegistry or discover_extensions.
validate_extension_spec success and failure cases.
Lazy imports remaining unloaded during discovery.
Missing help/data resources and missing required dependencies.
Duplicate action or pop_* names when relevant.
pop_* return values, return_com=True, history strings, and console auto-store behavior.
GUI visual parity and user-flow behavior for GUI extensions.
Wheel or source distribution contents when packaged resources matter.

Run the narrowest relevant pytest files first, then ./pre-commit.py --fix, focused tests, and broader ruff/format/ty checks when the change affects imports, docs, packaging, or public APIs.

Distribution Decision

Local/private lab only: uv add --editable /path/to/eegprep-ext-foo.
GitHub-only: uv add git+https://github.com/lab/eegprep-ext-foo; pin tags or commits for reproducibility.
PyPI: uv add eegprep-ext-foo after normal package publishing.
Private index: uv add --index https://packages.lab.example/simple eegprep-ext-foo or uv add --default-index ... when the private index replaces PyPI.

EEGPrep does not host extension zips, install arbitrary archives, or treat curation as scientific endorsement.

Catalog Submission

Before proposing curation, verify:

The package installs from a standard path and the entry point returns a valid ExtensionSpec.
The extension passes validation in a clean environment.
Menus, help, package data, pop_* history, and console behavior are covered.
Optional dependencies and large files are documented and do not block startup.
Experimental status is explicit in docs and metadata.
Governance/catalog checks pass when the extension is intended for catalog submission.