dsh-wakatime

When an AI agent inside DeepSeek Harness writes, edits, and swaps code at speed, those changes routinely slip past your personal coding statistics. dsh-wakatime bridges the file-operation events in dsh's agent loop to WakaTime, so AI-assisted coding lands in the same clear ledger as your hand-typed lines.

The plugin is built around precision and liveness. It reads the durable meta from the fs tools—resolved entity paths and per-hunk diffs—to count only genuinely added or modified lines and exclude context. During long streaming turns it fires near-real-time heartbeats driven by agent/status and assistant-stream events rather than waiting for durable settlement. A once-per-minute, per-project rate cap is persisted to a state file shared by all parallel dsh processes, preventing double counting. Batched edits collapse into a single wakatime-cli call via --extra-heartbeats, and the entire plugin carries zero runtime dependencies beyond Node builtins and the peer packages the host already provides.

If you drive AI coding through dsh's web, tui, or headless profiles and already track manual development on WakaTime, this plugin is the missing link. After installing it, every line the AI changes, every file it reads, and every streaming generation flows quietly into the WakaTime dashboard you already know—no extra configuration, no daemon to babysit.