temrb/agent-configs：可维护、可复用的 agent 资产仓库

Maintainable, reusable agent assets for supported clients. The filesystem is the index: browse agents/ to discover the current entries.

Principles

One canonical source: edit an asset in its canonical collection rather than a bundled or installed copy.
Explicit distribution: generated plugin assets are reproducible and committed so packages can be used independently.
Client-neutral core: common skill, plugin, and build policies apply across clients; client-specific configuration and adapters belong in client-specific locations.
Documented compatibility: a portable file layout is not a promise that every agent client installs or executes it identically.
Safe maintenance: validate sources, generated artifacts, and ownership before release; never commit runtime credentials or state.

Collections

Collection	Purpose	Canonical location
Skills	Reusable Agent Skills	`agents/skills/<name>/`
Plugins	Distributable plugin packages with optional bundled components	`agents/plugins/<name>/`
Configurations	Client-native configuration and companion assets	`agents/configs/<platform>/`

Each collection README defines its asset contract, installation or consumption boundary, and extension procedure. Client-specific instructions, including Codex, Meta Muse Code, OpenCode, and Prefine installation, live with their entries.

Maintain

Python 3.11+ is required. From the repository root:

python3 scripts/repo.py sync
python3 scripts/repo.py sync --check
python3 scripts/repo.py validate
python3 -B -m unittest discover -s tests
python3 -B scripts/check_docs.py

Edit canonical sources first. sync rebuilds declared generated outputs; sync --check detects drift without writing; validate checks supported asset contracts. The tests exercise rendering, ownership, and validation. Generated contents and their .generated.json inventories are committed together.

See Build and maintenance for recipes, ownership, recovery, adoption, and retirement; see Standards and compatibility for terminology, external specifications, and repository-specific policy.

Adding entries

Add a skill, plugin, or native configuration beneath its collection using that collection's README. No manually maintained root catalog is needed. New generated collections are registered in repo-build.json; new asset types or validators also require explicit code and test coverage. See extension boundaries.

Releases

Review each consumer of changed canonical inputs, synchronize, validate, run the tests, and commit generated outputs and ownership inventories. Plugins use semantic versions and immutable tags; standalone skills use Git commits as release identities. See release policy.