Scored across 38 tools
There are many skill-related tools with heavily overlapping purposes, such as create_skill, create_skill_draft, generate_skill, import_skill, adopt_skill, and promote_skill_recommendation. An agent would struggle to choose the correct tool without deep domain knowledge, and the boundaries between latent tracking, proposals, recommendations, and canonical skill creation are not obvious.
Most tools follow a consistent snake_case verb_noun pattern, like list_facts, purge_fact, validate_skill, and rename_skill. There are only a few deviations such as bootstrap, status, context, doctor, host_setup, and host_check, which are still readable but break the otherwise predictable convention.
38 tools is excessive for a single server, especially when roughly two dozen are dedicated to skill lifecycle management. This creates unnecessary selection burden and overlaps; the server would benefit from consolidation or splitting into focused sub-servers.
The server covers a broad range of memory, snapshot, audit, host, and skill lifecycle workflows with surprising depth. However, there are minor gaps like updating an existing memory fact, deleting a canonical skill entirely, or rolling back to a specific snapshot rather than only the latest.