AI Agent Hub
Back to plugins
🖥️

dsh-rich-tracking

Client Updated 2026.08.27

Run the following command in DeepSeek Harness:

dsh plugin install savageops/dsh-rich-tracking

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install savageops/dsh-rich-tracking inside your DeepSeek Harness environment to install this plugin; the full source repository is available at https://github.com/savageops/dsh-rich-tracking .

About this plugin

In long-horizon agent sessions, progress often lives only in a per-turn todo list that resets every step: there is no persistent cross-turn board, percentages lack artifact-level evidence, and operators have no lever to steer mid-mission. dsh-rich-tracking plants a durable mission scoreboard in the conversation dock so that every row is anchored to real artifact truth rather than narrative.

Each row percent is derived from verifiable acceptance items — plan checkboxes, landed receipts, verified readbacks — and the schema validation rejects any percent that lacks evidence. Checkpoints are captured by the host directly from git branch, HEAD, and dirty state, so the model never types git facts by hand. Operators wield whip-actions (pursue, delegate, scout, align, checkpoint, dismiss, play) that land as attributed instructions in the agent's next step; the /track command forces a full ledger sync, and a staleness nudge fires after eight steps or six thousand tokens without a tracking_write.

Built for operators running multi-step, multi-session coding or research missions through DeepSeek Harness who need an honest, evidence-bound progress view and direct steering levers that remove the need to keep asking where things stand.

Use Cases

  • Track per-row completion honestly across multi-session coding missions with an evidence-bound scoreboard
  • Enforce percent honesty via align re-derivation and staleness nudges anchored to artifact truth
  • Delegate rows to background subagents and fold their commits and test receipts back into the parent board automatically

Best For

  • Operators running multi-step coding or research missions through DeepSeek Harness
  • Team leads who need an auditable, replayable progress view for long-horizon agent tasks
  • Developers who prefer whip-action steering over repeatedly asking the agent where things stand