| name | modules |
| description | Umbrella catalog of the reference Python modules the spec pipeline copies verbatim into a target repo during init/scaffold (shared domain primitives, persistence context packages, REST serializers). Each group is a sibling folder of this file. Load to resolve the modules directory and copy a group by path. |
| user-invocable | false |
Reference modules (spec-core)
The spec pipeline ships a set of runtime contract modules — real Python source the generated code imports (Entity, ValueObject, guards, UnitOfWork, QueryContext, DatabaseSession, the serializer base). They are not generated per-aggregate; they are copied verbatim into the target repo once, by each layer's init/scaffold flow.
They are homed here in spec-core — the shared foundation every spec plugin already depends on — so there is one canonical source with one resolvable path, instead of each plugin resolving its own bundle via brittle find ~/.claude/plugins globs or impossible "walk up to a plugins/ ancestor" heuristics. Those heuristics do not work from an agent's or skill's runtime Bash (the ${CLAUDE_PLUGIN_ROOT} variable is not exposed there); the umbrella mechanism does, because loading this skill reveals its own directory.
Each group is a supporting folder of real .py files (not an index.md doc) sibling to this SKILL.md. The files are never auto-loaded into context — consumers copy them to disk on demand.
Resolution rule
A consumer resolves <modules_dir> as the directory containing this SKILL.md — its loaded context (frontmatter auto-load for agents, or the Skill tool for skills) reveals its location. A group named <group> is then the folder <modules_dir>/<group>/.
Copy a group with a directory-tree copy from the resolved absolute path:
cp -r <modules_dir>/<group> <destination>
Be idempotent: never overwrite a file that already exists at the destination; copy only what is missing. Preserve contents byte-for-byte. A group name with no matching folder is an error — report it loudly; never skip it silently.
Catalog
| Group | Folder | Copied by | Contents |
|---|
| shared | shared/ | domain-spec:init-domain | domain primitives — entity.py, value_object.py, entity_id.py, event.py, command.py, clock.py, extended_enum.py, exceptions.py, pagination/result-set DTOs, and the nested guards/ package |
| database_session | database_session/ | persistence-spec:database-session-scaffolder | __init__.py, constants.py, database_session.py |
| unit_of_work | unit_of_work/ | persistence-spec:context-package-scaffolder (unit_of_work axis) | __init__.py, abstract_unit_of_work.py, sql_alchemy_unit_of_work.py |
| query_context | query_context/ | persistence-spec:context-package-scaffolder (query_context axis) | __init__.py, abstract_query_context.py, sql_alchemy_query_context.py |
| serializers | serializers/ | rest-api-spec:serializers-copier | error.py, configured_base_serializer.py, json_utils.py |
Some consumers patch a freshly-copied file after the copy (e.g. context-package-scaffolder rewrites the DatabaseSession import placeholder; serializers-copier regenerates the aggregator __init__.py). That post-copy work is owned by the consumer, not this catalog. The sql_alchemy_* files under unit_of_work/query_context intentionally ship with a # Add DatabaseSession import there placeholder line for that reason.
