Scored across 14 tools
Several tools overlap around reading the .faf: faf_score, faf_doctor, faf_trust, and faf_context all read/scorer the same file with subtle differences (score vs. diagnose vs. attest vs. active project). faf_sync and faf_tri_sync both write faf-managed blocks and only differ by target file, and faf_init/faf_auto/faf_go all create or fill slots. Descriptions do clarify the distinctions, but an agent could easily pick the wrong read or write tool.
All tools use the faf_ prefix with a short lowercase verb or concept (faf_about, faf_score, faf_init, faf_sync, faf_go, faf_etch, faf_recall). The pattern is consistent and readable, though the second token is sometimes a verb and sometimes a noun (about, context, bench), so it is not a strict verb_noun convention.
14 tools is within the reasonable 3-15 range for a server covering the full .faf lifecycle (init, fill, score, diagnose, sync, memory, attest, bench). It is on the heavier side but each tool has a distinct conceptual role, so the count is defensible.
The surface covers the .faf lifecycle well: create (faf_init), auto-fill and human answers (faf_auto, faf_go), read/score/validate (faf_score, faf_doctor, faf_trust), write to editor files (faf_sync, faf_tri_sync), session hook setup (faf_setup), and memory (faf_etch, faf_recall). Minor gaps exist, e.g. no explicit delete/remove for project.faf or memories, but core workflows are complete.