Related Servers

Descriptions are unusually thorough and explicitly cross-reference the right sibling (e.g. judge_objective vs judge_objectives vs judge_imported_controls vs recompute_verdicts; refine_control vs remap_control vs regenerate_controls vs apply_control_changeset), which prevents most misselection. However, there are large overlapping clusters (start/pause/resume/discard control builds; get_controls/get_control_objectives/get_verification_report/get_sufficiency) and a real terminology collision where 'groups' means tags in list_groups/get_group/create_group while mitigation/assumption/satisfaction groups mean something else entirely. An agent can still pick correctly by reading carefully, but the boundaries are genuinely crowded.

Nearly everything follows a predictable snake_case verb_noun pattern (add_asset, edit_attacker, remove_entity, list_findings, get_threat_model, submit_assertions). A few stative/odd names (assess_model, recompute_verdicts, judge_objective) and the decision to name tag tools 'group' while calling them tags in prose are minor deviations, but the style is internally consistent throughout.

128 tools is an extreme count that far exceeds what an agent can reliably select from, and the surface contains visible redundancy (three judge_* tools, four control-build lifecycle tools, multiple verdict/report readers). Even granting the platform spans threat modeling, assurance, compliance, functional testing and composition, the number is a mismatch for practical tool routing.

Coverage is essentially exhaustive: full CRUD/lifecycle for models and every core entity (assets, attackers, components, trust boundaries, assumptions), control build/undo/versioning, assertions and sufficiency, findings, mitigation and satisfaction groups, risk acceptances and dispositions, compliance frameworks, functional tests, proposals/decisions, and composition lift/split. No obvious domain operation is missing, and import/export/archive round-trips close the loop.