danbiocchi

Context engine that replaces the LLM-written compaction summary with a deterministic projection of the conversation: requests, commands and exit codes, errors, fixes and files read or changed, each line referencing the original message for recall via session_search. No model call on the normal path. Disclosure — no network calls, shell commands or credentials; reads Hermes's config.yaml through Hermes's own loader; subclasses ContextCompressor and relies on its private methods, falling back to the stock compressor if they are missing.