livery.extensions.claude._agent¶
livery.extensions.claude._agent ¶
What Claude Code reads in a workspace, answered for the workshop's engine.
The extension's extension.toml names these functions as its
[fragments] renders: the workshop calls each with the workspace
root while a workspace lists the extension, and writes what it
answers. CLAUDE.md is committed; .workshop/fragments/ and
.claude/ belong to the checkout alone.
Functions:
| Name | Description |
|---|---|
entry_file |
|
guidance_files |
|
claude_files |
|
Attributes:
| Name | Type | Description |
|---|---|---|
DELIVERED |
|
|
OWN |
|
|
STUB_HEADER |
The |
STUB_HEADER
module-attribute
¶
STUB_HEADER = "<!-- Managed by `{prog} sync`: one import per fragment, in section\n order, the repository's own fragments/ after them, then its\n CLAUDE.project.md, which always wins. Edit CLAUDE.project.md,\n never this file. -->\n"
The CLAUDE.md stub's header, formatted with the runner's name.
entry_file ¶
CLAUDE.md: an import per delivered fragment, the own ones, the project's.
Reads the composition guidance_files made just before it in the
delivery; rendered on its own, it composes for itself.
Source code in packages/extensions/claude/src/livery/extensions/claude/_agent.py
guidance_files ¶
.workshop/fragments/: each fragment the agent reads, by delivered name.
Declared before the entry file, so the entry file reads this composition instead of making its own.
Source code in packages/extensions/claude/src/livery/extensions/claude/_agent.py
claude_files ¶
.claude/: each listed extension's skills and hooks, and one settings.json.
A skill or a hook is a link into the shipping extension's content, and a later extension's entry of one name takes an earlier one's place. The settings are a copy, which the agent's tooling may edit, and at most one extension ships them.