Claude Code is docket's reference harness, and everything in Installing docket and Models applies to it directly. Two Claude-Code-specific details are worth calling out.
Forking preserves the pin but hides the run. As the How the pin survives a direct invocation
table in Models shows, a directly-invoked forked skill returns as
completed (forked execution), with no box to drill into in the TUI. A forked run is not lost, only
unobservable there: Claude Code still writes its full transcript to
~/.claude/projects/<project-slug>/<session-id>/subagents/agent-<id>.jsonl. Treat that path as an
observed internal, not an interface — it was accurate on Claude Code 2.1.207, it may move, and
docket depends on it for nothing. When you want to watch a long run live, dispatch it with
@docket-<name> instead, which routes through a real, drillable dispatch.
Restart after changing an agent or a skill. Skills and agents are registered at process start. After an install, or after you edit a skill's frontmatter, an already-open session keeps running the old definitions — so a freshly-added fork appears to do nothing, and a healthy pin looks broken. Restart the harness process (a new session — clearing the context is not enough) and re-invoke.