Skip to content

livery-extensions-claude

What Claude Code reads in a workspace. A workspace turns it on by listing the extension:

[workspace]
extensions = ["claude"]

fm sync then writes three things:

  • CLAUDE.md, committed: an import line for each guidance fragment in section order, the repository's own fragments/ after them, then CLAUDE.project.md, the repository's own facts. Edit CLAUDE.project.md, never CLAUDE.md.
  • .workshop/fragments/, this checkout's alone: the guidance every listed extension ships or renders, for the agent.
  • .claude/, this checkout's alone: a link to each skill and hook every listed extension ships under content/skills/ and content/hooks/, and a copy of the one settings.json. A copy the agent's tooling edits is a local override, kept and named.

The settings wire Claude Code's hooks to the extension's verbs through one shim, .claude/hooks/fm-hook.sh:

  • fm hooks.pre-bash refuses a footman command piped into tail or head, and a push of a branch that conflicts with origin/main.
  • fm hooks.post-edit runs the gate's fixers over the edited file.
  • fm hooks.stop refuses to end a turn whose last command was a failed fm verdict, once.

A project born while the extension is listed starts with a CLAUDE.project.md; fm new.project lists it only when asked. A workspace that does not list it has no CLAUDE.md and no .claude/.