In the plugin ecosystem of DeepSeek Harness, maintaining data privacy while enabling agent self-evolution is a common requirement. The Missher Evolution plugin provides a solution. It adapts the MSE SDK to preserve bounded workflow experiences and preference management while ensuring data isolation.

Core Positioning

This is a management and security plugin maintained by Missher12, designed to provide privacy-bounded self-improvement capabilities for DeepSeek Harness. It does not directly train model weights or modify Harness itself; instead, it adapts the MSE SDK to perform bounded learning and management of workflows, preferences, and results in a local environment.

Core Features

  • MSE SDK adaptation: Adapts MSE SDK for Harness to enable Harness-side integration.
  • Bounded experience preservation: Preserves bounded workflow experiences, preferences, and verifiable results.
  • Intelligent checks and repair: Supports source-date and unknown-value checks, and preserves metadata in failure/repair scenarios. The maintenance phase uses AI budgeting or deterministic fallbacks.
  • Brain Hub service: Provides the Brain Hub service to handle lifecycle events.
  • Data isolation: Uses SHA-256 scoped hashes to implement data isolation.
  • Local state management: Manages local directory state under $DSH_HOME/missher-evolution/.

Installation and Enablement

Before use, ensure the host environment meets the requirements: DeepSeek Harness Desktop >= 0.3.8.

Installation Steps

  1. Download the 0.7.0 tarball installer from GitHub Releases.
  2. Install it using the DSH client command:
dsh plugin --profile <profile> add https://github.com/Missher12/Missher-DSH-Evolution/releases/download/v0.7.0/dsh-missher-evolution-0.7.0.tgz

Enablement and Verification

After installation, open the Harness settings interface and select “Missher Evolution” to check the status and enable learning. To confirm that the plugin has been loaded correctly, run the following command:

dsh --profile <profile> --dump-config

The output should include the dsh-missher-evolution bundle layer and the enabled missher-evolution host entry.

Typical Usage

  • Status check: Go to Settings -> Missher Evolution and view the current learning status and rules.
  • Rule reset: Find the “Reset” option in settings to clear learned rules (preferences are retained).
  • Data management: The plugin’s local state is stored under the $DSH_HOME/missher-evolution/ directory.

Notes

  • Version lineage: This is part of the MSE 0.7 product line and is maintained separately from the Missher-MSE-Learning alpha product line.
  • Data privacy: Data is fully local and does not include raw user messages, credentials, or API keys.
  • Version limitation: Version 0.7.0 does not expose per-rule reviewRule approval/revocation or the restore remote interface.
  • Profile sharing: Different profiles under the same DSH_HOME share evolution state; the profile itself cannot isolate this data.