[infra] Scaffold template for contextweaver plugin projects

[dgenio]

Problem

Developers wanting to create contextweaver plugins (custom retrievers, event hooks, store backends, summarizers) have no template or scaffold. They must reverse-engineer the protocol definitions, entry-point configuration, and testing patterns from the main repo. This friction discourages ecosystem contributions.

Proposal

Add a contextweaver init --plugin CLI command that scaffolds a minimal plugin project:

python -m contextweaver init --plugin my-retriever

Generates:

contextweaver-my-retriever/
├── pyproject.toml          # entry-point pre-configured
├── src/
│   └── contextweaver_my_retriever/
│       ├── __init__.py
│       └── retriever.py    # stub implementing Retriever protocol
├── tests/
│   └── test_retriever.py   # protocol conformance test
├── README.md               # usage instructions
└── .github/
    └── workflows/
        └── ci.yml          # basic CI (lint + test)

Plugin types supported

The --type flag selects which protocol to scaffold:

Type	Protocol	Stub
retriever (default)	Retriever	score() method
hook	EventHook	on_event() method
store	EventLog / ArtifactStore	CRUD methods
summarizer	Summarizer	summarize() method

Template content

• pyproject.toml with correct [project.entry-points] section
• Stub class implementing the chosen protocol
• Conformance test using isinstance() and basic smoke test
• README with installation and registration instructions

Acceptance Criteria

• python -m contextweaver init --plugin <name> creates a valid project directory
• --type flag supports retriever, hook, store, summarizer
• Generated pyproject.toml includes correct entry-point configuration
• Protocol stub passes type checking with mypy --strict
• Generated tests pass with pytest out of the box
• Generated README explains how to install and register the plugin
• Templates are stored as importlib.resources data (not hardcoded strings)

Implementation Notes

• Use importlib.resources to load templates from a templates/ package directory
• String interpolation (not Jinja) to avoid adding dependencies
• Add the init --plugin subcommand to __main__.py
• Keep templates minimal — users can expand as needed

File Paths

• src/contextweaver/__main__.py (edit — add init subcommand)
• src/contextweaver/templates/ (new package with template files)
• tests/test_cli.py (edit — test init subcommand)

Cross-references

• Depends on [routing] Pluggable retrieval backend with entry-point discovery #121 (pluggable retrieval backend — defines Retriever protocol)
• Related to [routing] Add EngineRegistry with pluggable Retriever, Reranker, and ClusteringEngine protocols #47 (EngineRegistry — plugin discovery)

[dgenio]

🔗 Dependency Note

This issue depends on #47 (retrieval architecture — Retriever/Reranker/ClusteringEngine protocols).

The real-world retrieval eval here requires the retrieval protocol layer from #47 to be implemented first.

Added during backlog hygiene review.