dsh-cap-profile

The more sessions you run, the harder it becomes to answer a simple question: which model is rock-solid on certain tasks and which one keeps tripping up on others. Scrolling through raw logs is slow; manual tallies inevitably miss patterns. dsh-cap-profile turns DSH (DeepSeek Harness) session history into a capability profile panel that shows, at a glance, each model s session count, tool-call volume, error rate, most-relied-upon tools, and highest-frequency error signatures.

The panel goes well beyond a single-model view. Select up to four models to enter a side-by-side comparison: convergent metrics auto-highlighted, a tool Top-10 matrix, top-5 recurring errors, and daily call/error trend mini-bars all laid out for quick differential reading. Time-range filters cover all-time, last 7 / 30 / 90 days, today, and yesterday, anchored to the latest date in the data so system clock drift is a non-issue.

Under the hood it is strictly read-only and never mutates session data. A per-file mtime baseline drives a 60-second incremental and 24-hour full refresh cycle in the background; if the initial scan is still running, the route falls back to the previous cache or mock data without blocking. Zero runtime dependencies, pure Node.js. If you run heavy DSH sessions daily and want to quickly gauge how models differ across task types, this panel saves you from hours of log-digging.