dsh_token_usage

After a few days of living in DeepSeek Harness, you probably start wondering where the token spend actually went, which model is burning the most, and whether cache hits are even being counted. The problem is that DSH ships no built-in usage view -- session logs sit in compressed local files, and pulling numbers out means writing your own scripts.

dsh_token_usage keeps things simple. It reads the session logs under the sessions directory, aggregates the token usage that providers report over the last 30 days, and drops a Usage entry just above the Settings button in the sidebar. Click it and you get a heatmap plus a per-model breakdown. The heatmap makes it obvious which days were heavy and which were quiet -- white-to-blue in light mode, black-to-blue in dark. The model table lays out input, output, cache reads, cache writes, and reasoning tokens in separate columns; the headline number is new tokens (input + output + cache writes), while cache reads and reasoning are listed on their own rather than folded in. Older sessions that carry no usage field are skipped entirely; the plugin never estimates from character counts.

The whole thing is a standalone bundle: no DSH source modifications, no sidebar replacement, no database. Install it, restart, and the entry appears. It is aimed at everyday DSH users who want a low-effort way to track token spend, compare usage across models, and keep an eye on cache hit rates versus reasoning overhead.