Knocklog

Each tool targets a clearly different object: classify_request analyzes a single live request, list_known_agents returns reference directory data, and scan_site audits an external website. Although classification draws on the same agent knowledge as the directory, the action and input shape are distinct enough that an agent would not misselect.

All three names use a consistent snake_case verb_noun pattern (classify_request, list_known_agents, scan_site), with clear verbs indicating the operation. No mixed conventions or vague names.

Three tools is on the thin side but each maps to a genuine, non-redundant workflow in a narrow domain. The set is focused rather than padded, though it sits at the lower boundary of what feels comfortably complete.

The core lifecycle — identify a request, look up known agents, and audit a site — is covered with no dead ends. Minor gaps exist: there is no batch classification, no retrieval of past classification results/logs, and no per-agent detail lookup beyond the directory listing.