Skip to main content
Agent-native marketplaces, registries, plugin repositories, and sources are surfaces that agent ecosystems expose to their users. Gandalf browses and inspects them through the Console Marketplace tab, but it never owns, certifies, or replaces the catalog itself.

​
What Gandalf shows

  • Codex: managed plugin skill inventory and source-backed entries where discovered.
  • Claude Code: marketplace source metadata plus installed and source-backed entries where discovered.
The Console renders these as unified inventory rows next to your installed skills, hooks, MCP servers, and plugins.

​
What actions are available

Browsing and inspection are always available. Mutating actions are gated by providers: Any unavailable action shows a concrete reason in the TUI rather than pretending to run.

​
Marketplace-originated Review Actions

A Marketplace-originated Review Action starts from an agent-native source entry. The first safe version does not mutate: it produces reviewed setup guidance or source-backed context. When a concrete provider exists (as with Claude Code plugin install), the same flow can preview and then execute a real change. Sources are treated as untrusted display data, so text is never executed at browse time.

​
Trust boundaries

  • Gandalf does not execute plugin, skill, hook, or MCP commands during scan or browse.
  • Source data is revalidated before any Marketplace-originated Review Action completes.
  • Install and rollback rely on the same content-backed snapshot safety path as restore.
See the Support Matrix for the exact current state per agent.