AI Agent Hub
Back to plugins
🖥️

dsh-compaction-tune

Client Updated 2026.09.01

Run the following command in DeepSeek Harness:

dsh plugin install kuanfu0430/dsh-compaction-tune

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

Run dsh plugin install kuanfu0430/dsh-compaction-tune in your DeepSeek Harness terminal to install; source code at https://github.com/kuanfu0430/dsh-compaction-tune

About this plugin

The stock compaction policy in DeepSeek Harness (@deepseek-ai/dsh-compaction-basic) ships with a static ratio pinned in the plugin YAML (default 0.8) and offers no in-conversation way to tweak it. dsh-compaction-tune drops a live chip into the lower-right of the composer, letting you set a compression trigger as a percentage (30–80%) or an absolute token count (64k–832k). The value is written to your local settings file and takes effect on the very next agent step—no restart required.

Under the hood the chip cooperates with the official compaction backend: once consumed tokens cross your threshold while still below the 80% safety cap, the plugin calls compactIfNeeded to compress proactively. If your threshold sits at the 80% ceiling, the official listener remains the sole trigger. When the optional anchored backend is loaded, this plugin simply exposes its policy service and registers no second trigger, preventing double compaction. Child sessions such as sidebar branches never inherit the main-chat threshold and retain the backend default.

Ideal for DeepSeek Harness users who juggle long and short context tasks, want per-task fine-grained control over when compaction fires, and prefer not to hand-edit the official plugin YAML. Note: installing this plugin alone does not add a compaction engine; your web profile must already ship compaction-basic or a compatible backend for the chip to actually trigger a compression pass.

Use Cases

  • Adjust compaction trigger mid-conversation without restarting
  • Quickly switch between percent and token thresholds across tasks
  • Skip manual YAML edits and dsh web restarts for every policy change

Best For

  • DSH users who handle long contexts and need on-demand compaction tuning
  • Developers who want per-task compaction boundaries without editing official plugin YAML
  • Light-use web profile users already shipping compaction-basic