AI Agent Hub
Back to plugins
dsh-task-capsule preview

dsh-task-capsule

Client Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install Chance-Wu/dsh-task-capsule

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

Run dsh plugin install Chance-Wu/dsh-task-capsule in your DeepSeek Harness terminal; the source repository is https://github.com/Chance-Wu/dsh-task-capsule.

About this plugin

When a multi-step task runs in DeepSeek Harness, the chat stream keeps scrolling and tool cards pile up, making it hard to glance at what is happening, how far along it is, and how long it has taken without scrolling back. dsh-task-capsule distills the entire execution into a compact, always-visible pill in the session header: a status dot, task counters, and an MM:SS timer, with a breathing animation that distinguishes running from waiting at a glance.

Click the pill to expand a liquid-glass floating panel that reveals the current operation, file-change statistics, sub-agent delegation tree with per-agent progress, per-task duration bars, and a persistent recent-task buffer showing success rate, average time, and one-click retry. The panel is portal-anchored so it never clips against scroll containers, auto-flips when space is tight, and handles focus cycling. Tasks auto-expand on start and auto-collapse when idle or complete; failures stay visible for quick triage.

If your Harness workflow involves long-running tasks, multi-agent delegation, or you simply do not want to interrupt your reading by scrolling through history to check progress, this plugin is built for that: minimal noise, zero distraction, always know where you stand.

Screenshots

Use Cases

  • Checking long-task progress without scrolling through chat
  • Tracking multiple sub-agent delegations at a glance
  • Reviewing task history with success rate and average duration

Best For

  • Harness users who frequently run multi-step long tasks
  • Developers whose workflows rely on sub-agent delegation
  • Users who read long outputs and prefer zero distraction