Scored across 65 tools
Tools cover many overlapping domains (CPU, memory, disk, network, process, Docker, etc.), but descriptions explicitly cross-reference alternatives (e.g., 'for X use Y'), making most selections clear. A few clusters such as process info/tree/fds and disk usage/inode/block devices still require careful reading to avoid misselection.
Almost all tool names use snake_case with a consistent get_<noun> pattern. The only exceptions are ping_host and resolve_dns, which use action verbs instead of get_, a minor deviation but still readable and consistent in casing.
With 65 tools, the surface is far beyond the typical well-scoped range (3-15) and exceeds the 50+ threshold for extreme mismatch. Even for a broad Linux diagnostics server, the set includes overlapping utilities (e.g., multiple process, disk, and network tools) that could be consolidated or scoped more tightly.
The read-only diagnostic surface is exceptionally comprehensive, covering system info, hardware, storage, network, Docker, systemd, security, logs, users, packages, and processes. Minor gaps exist (e.g., no sysctl/kernel parameter read, no partition table detail, no arbitrary file-content inspection), but agents can work around most of these.