Related Servers

Many tools carry explicit routing prose ('use X for..., use Y for...'), which genuinely helps separate lookalikes such as task_show vs agent_task_status vs supervisor_loop_status and task_list vs review_queue vs plan_snapshot. But the set still has real overlap clusters: completion_inbox vs dashboard_snapshot vs task_review_queue, codex_handoff vs codex_handoff_markdown vs review_summarize, needfix_archive vs dashboard_needfix_transition vs needfix_update, and needfix_add vs manager_needfix_draft. At 117 tools the reader must hold a large disambiguation graph in working memory even though each individual pair is documented.

The dominant style is snake_case, and the manager_* family (manager_kb_get/search/related, manager_recipe_list/show/run) is internally consistent. However the set mixes an aiworkhub_ prefix on most tools with unprefixed needfix_* and roadmap_* families, and mixes verb_noun names (task_create, needfix_archive) with bare noun names (review_packet, completion_inbox, dashboard_snapshot, task_health, task_collision_guard, task_review_queue). It is readable but not predictable from name alone.

The platform genuinely spans many subsystems (task lifecycle, review, NeedFix, roadmap, skills, recipes, KB/memory/context graph, source graph, dispatcher, storage), so a large surface is partly earned. Still, 117 tools is far past what an agent can reliably select from, and several near-duplicate or facet-of-a-larger-tool entries (needfix_count, codex_handoff_markdown, dashboard_snapshot) look like candidates for folding into callers that already exist. This is over-scoped for reliable tool choice even if few tools are individually trivial.

Coverage is unusually thorough: create/amend/list/show/archive/supersede/supersede-recover/retry/reroute for cards, full review evidence plus accept/preview/reject/finding-dispose/hold-resolve, and complete lifecycle sets for NeedFix (add/update/archive/restore/convert/events/count), roadmap, skills, recipes, KB, AI memory and context graph. Preview/commit/status/health/diagnose/doctor companion tools mean few workflows dead-end, and read-only previews exist alongside every gated write. No material gap is apparent for the stated orchestration purpose.