Scored across 23 tools
The descriptions explicitly differentiate overlapping tools, such as audit_log vs semantic_recall for time-ordered vs topic search, and decision_check vs decision_review vs decision_log for different decision operations. However, several read-only state tools (operational_state_check, entity_read, project_evidence_scan) and multiple decision/memory check tools could still cause momentary confusion for an agent without careful reading.
All tools use snake_case with a consistent domain-first convention (e.g., decision_check, entity_read, plan_advance, memory_commit). Minor deviations like wrap_auto and action_guard still follow the same snake_case style, so there is no mixed convention.
At 23 tools, the set is on the heavy side for the server's purpose, though each tool targets a distinct operation within memory, decisions, plans, wrap, risk, and audit. The count is borderline per calibration and could likely be consolidated (e.g., plan and decision families) without losing capability.
The surface covers a wide lifecycle: decisions (log, check, review, refresh), entities (read, update), plans (set, add, advance, read), memory (check, commit, recall, audit), wrap, risk assessment, and evidence scanning. Minor gaps remain, such as no explicit entity/plan deletion or a direct list-all-decisions tool, but agents can work around these via entity_read and decision_review.