dsh-file-mount

In LLM workflows, repeatedly reading files often creates a hidden cost. Logs, configs, code, and documentation can keep filling the context window even when only a small part has changed. dsh-file-mount addresses this by tracking which line ranges of each file have already entered the model context, then avoiding redundant injection on later reads and supplying only the missing or newly changed content.

Its core strength is a file-level context ledger rather than a simple cache. When a file changes on disk, the plugin can compare content line by line and mount only the updated portions, which is especially useful for appended logs. It also recognizes files previously touched by the model and can reuse known context more intelligently. When freshness matters, file_mount_forget lets the model explicitly force a re-read. The bundled dashboard visualizes mounted ranges, freshness, overwrites, savings, search, and sorting, turning token control into something inspectable instead of opaque.

This plugin is best suited for DeepSeek Harness users who repeatedly work with codebases, logs, documentation, configuration files, or long-running conversations. It is especially valuable for people who care about context hygiene, token savings, and observability, and who want a practical way to prevent stale content from crowding out useful context.