Human Capability Registry

Several tools overlap by design: summon_expert explicitly re-implements both request_review (mode=verify) and join_expert (mode=live), so an agent must decide between a specialized tool and its generic alias. The trio search_human/ask_human/call_human share confusingly similar names with distinct purposes, and trust_lookup/verify_human/ask_card all read registry trust data. Descriptions do a good job steering choices, but the boundaries are not clean.

All 16 tools use consistent snake_case, which is a strong baseline. However the verb_noun convention is not uniform: most are verb_noun (ask_card, search_human, call_human), but room_close/room_wait invert to noun_verb and deliver/status are single tokens, so the pattern is mostly—not perfectly—predictable.

16 tools sit at the high edge of the comfortable range, but the domain genuinely spans discovery, hiring, payment/status, live rooms, verification, public records and routing, so most earn their place. The main bloat is summon_expert duplicating join_expert and request_review; without that overlap the count would be cleaner.

The surface covers the full arc from routing and search through hiring, checkout, delivery, live rooms, verification and off-platform records. Notable gaps remain: there is no tool to release/finalize payment despite customer_token being described as needed for 'release', and no dispute or card-claim operation even though status references a 'disputed' state and cards can be 'claimed'.