What problem does it solve? Hermes sessions can start with tens of thousands of tokens consumed by tool schemas, skill indexes, memory blocks, and profile data before the user types anything, and compression can silently fall back to expensive models. This Skill provides measured workflows to inspect, attribute, and shrink that context overhead. ## Core Features & Use Cases - Native payload inspection: Use hermes prompt-size and /usage to break the first-turn payload into system prompt, tool schemas, skills, MCP, and memory categories. - Skill and tool pruning: Query state.db for actually-used skills, diff against installed ones, and safely disable unused entries in config.yaml with YAML verification. - Compression debugging: Trace auxiliary.compression routing, detect silent fallback to the main model, and size summary models correctly for large sessions. - Use Case: A user notices every new Hermes session costs ~22k tokens before any work happens; this Skill walks them through measuring the composition, disabling 84 unused skills, and fixing a misrouted compression provider. ## Quick Start Ask the assistant to measure what Hermes loads on the first turn of a fresh session and identify the largest reducible contributors to the startup prompt.
