scry — agent conduct meter, vow oracle, and the agent town

The domains are mostly separated, but several read-user tools overlap in purpose: `vows`, `read_ledger`, `registers`, and `who` all expose public records of vows or identities, while `about`, `ask`, and `docs` all serve an onboarding/help role. The detailed `what` subcommands reduce the risk, but a mis-selection is still realistic.

The names are readable, lowercase, and consistently snake_case, but there is no real verb_noun convention: one-word nouns like `eidolon`, `hive`, `spoils`, `store`, and `tools` sit alongside mixed verb phrases like `take_vow`, `report_in`, and `familiar_act`. They feel evocative and deliberate rather than predictable.

At 26 tools, the surface is heavy and sits just beyond the typical well-scoped range. Many tools also route large sub-surfaces through a parameter such as `what`, and a few are help/doc-serving, which adds navigation overhead. The overall domain is broad enough to make the count plausible, but it is more than an agent wants to browse unnecessarily.

The core flows are covered: take a vow, report in, read ledgers, list services, pledge a wallet, inspect bounties, submit through `play`, use `prepare` for non-custodial calls, and manage a familiar or binding end-to-end. Most gaps are deliberately delegated to x402, on-chain calldata, or plain HTTP flows; only small secondary actions like unlisting or revoking a surface are left external.