@dsh-cc/command-plugin

@dsh-cc/command-plugin

English | 中文

Human-facing /plugin and /reload-plugins commands: list the mounted Claude Code plugins (name, root, and per-component load counts) and rescan the on-disk discovery roots to remount them live. The plugin registers two global commands through ctx.commands, so every composed command adapter finds and executes them without a model turn.

Command contract

Input	Result
/plugin	List each mounted plugin's manifest name, plugin root, and per-component counts (commands: N, agents: N, skills: N, …), annotating non-zero skipped/failed components.
/reload-plugins	Dispose all tracked mounts in reverse mount order, re-run discovery, and remount; report the fresh mount list plus any per-root failures.

Both read the optional ccPlugins service that cc-shell-glue mounts. A composition without the glue reports the seam gracefully instead of failing. Neither command consumes model tokens.

Composition

The producer injects commands. A custom app mounts the glue plus this plugin:

- id: cc-shell-glue
  name: '@dsh-cc/bundle-shell'
- id: command-plugin
  name: '@dsh-cc/command-plugin'

The ccPlugins seam is discovered via ctx at run time; it is not injected, so the command loads even when glue is absent (and then reports the missing seam).

Model Experience

The slash input and the direct outputs are absent from model requests and consume no model tokens. Outputs are pure reads of the mounted-plugin registry; presentation text is never logged.

Known Limitations and Deferred Work

• Plugin mount internals are glue-owned — the command only enumerates/rescans what ccPlugins already tracks; it cannot mount a plugin root that discovery skipped.
• No per-component detail — the report is counts plus reasons where the loader supplies them; verbatim component bodies are out of scope.