Related Servers

The core FSM `step` tool is a single entrypoint that dispatches to ~14 embedded actions, while the named tools (step, fork_at, fork_from_past, reset_session) all deal with session/FSM lifecycle and heavily overlap in purpose—all four are about navigating or resetting the state machine state. The `read_resource`/`list_resources` pair is distinct, but the boundary between step-driven control flow and fork/rewind/resume tools is genuinely blurry and would cause misselection.

The tool names use consistent snake_case and a Verb_Noun structure (read_resource, list_resources, reset_session, fork_at, fork_from_past), which is fairly consistent. However, `step` and `fork_at`/`fork_from_past` use dramatically different verb styles—`step` is vague while the others are specific—and there's no clear naming cue that these all orchestrate the same state machine. The embedded actions inside `step` are well-named, but the surface-level naming lacks a cohesive pattern.

Six tools is a well-scoped count for an orchestration server whose real surface is a state machine. The design deliberately centralizes the ~14 pipeline actions behind `step`, so six surface tools appropriately capture both the FSM orchestration (step, reset, forks) and resource access (read/list). This is reasonable for the stated purpose.

The FSM covers an impressively complete lifecycle: entry, intermediate pipeline stages, validation/fix loops, telemetry correlation, analysis, screening, and final reporting—plus reset and multiple resume/fork mechanisms. However, there are notable gaps: no tool exposes the FSM's current state/history/valid-next directly (only indirectly via `step` errors and `theodosia://history`), and no mechanism to inspect available code actions before stepping. The `read_resource` mechanism partially covers this, but it's not a first-class surface capability.