Scored across 18 tools
The set contains several near-sibling families — four assess_* variants, three evaluate_* variants, and a compare_code/compare_files pair — which is inherently risky, but each description explicitly states when to use it and, unusually, when NOT to (e.g. 'do not call this for one file', 'use topos_assess_worktree_change for in-place edits'). The remaining tools (coverage, depgraph, docs, preference_walk, pr_recap, inspect_code) are clearly separate concerns. A careless agent could still pick the wrong assess_* variant, but the guidance resolves most ambiguity.
All names use a uniform topos_ prefix plus snake_case, and nearly all follow a verb_noun shape (evaluate_file, assess_changeset, begin_refactor, calculate_coverage, generate_depgraph). A few are bare verbs or noun phrases (topos_refactor, topos_pr_recap, topos_preference_walk, topos_depgraph_status), which breaks the pattern slightly but remains readable and predictable.
At 18 tools this sits in the heavy/borderline band for a single analysis domain, and a chunk of the count comes from input-shape variants of the same operation (evaluate_code/file/project, compare_code/compare_files, four assess_* tools). The surface is coherent and each tool has a defensible role, but consolidation could cut the count meaningfully without losing capability.
The refactor lifecycle is well covered end to end: snapshot creation (begin_refactor), evaluation at snippet/file/project granularity, four diff/assessment modes, hotspot ranking, AST comparison, dependency-graph status/generation, coverage, and doc lookup. Gaps are minor — no way to list or delete stored snapshots, and no explicit configuration/preference-setting tool (preferences are only documented, not applied).