AI Agent Hub
Back to plugins
⚙️

dsh-progressive-tools

Workflow Updated 2026.08.27

Run the following command in DeepSeek Harness:

dsh plugin install everclear077/dsh-progressive-tools

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

Run dsh plugin install everclear077/dsh-progressive-tools in your terminal to add dsh-progressive-tools to your DeepSeek Harness profile; the plugin source is available at https://github.com/everclear077/dsh-progressive-tools .

About this plugin

Every tool definition a Harness profile registers is shipped on every single request, silently consuming input tokens and—worse—breaking context-cache prefix reuse the moment that list is mutated. For deployments that pull in a dozen plugins and hundreds of tools, most of those definitions are irrelevant to any given conversation yet still occupy the context window on every turn.

dsh-progressive-tools solves this with a stable-proxy discovery model: the first AgentLoop request carries only two lightweight tools (tool_search and tool_dispatch) while the full catalog lives in process memory. When the agent needs a capability, a deterministic BM25-style lexical search surfaces the exact definition without altering the top-level tool list, so the request prefix stays byte-stable across turns. A single search also expands an entire tool family, and a monotonic routing guard ensures deferred tools can only be executed through the dispatcher—preserving approval flows, sandbox policies, and argument validation for every real tool invoked.

The plugin is a strong fit for large Harness deployments that juggle multiple plugin bundles, skill bindings, and custom agent tools, and that want to cut per-request token overhead and improve cache hit rates without weakening existing security controls. A dynamic compatibility mode is also available for environments that require native tool definitions after activation.

Use Cases

  • A large number of registered tools inflates per-request input token cost
  • Preserving a byte-stable request prefix to maximize context cache reuse
  • Discovering and loading tool definitions on demand across multiple plugin bundles
  • Reducing the top-level tool surface while keeping approval, sandbox, and validation policies intact

Best For

  • Harness users running multiple plugin bundles, skill bindings, and custom agent tools
  • Deployment teams sensitive to token cost and context-cache hit rates who scale horizontally
  • Developers who want to reduce per-request payload without weakening existing security controls