Preface

The philosophy of DSH is “Everything is a plugin.” For AI agent developers, skills, MCP, and plugins are often scattered across different directories and configurations; manual registration is prone to omissions, and there is a lack of continuous metrics after calls are made, making it hard to judge whether a capability is worth keeping. dsh-genome puts these three types of capabilities into a unified management surface of registry / metrics / history, using a negative feedback servo loop and a progressive stability mechanism to make extensions observable, rollable, and stoppable.

What is it

dsh-genome is DSH’s capability transformation and assembly engine. The repository address is https://github.com/Elohia/dsh-genome, license is MIT; verified facts do not directly state the maintainer.

The core problem it solves is: how to make DSH’s skill / MCP / plugin libraries not just a one-time installation, but enter a management loop of continuous evaluation, decision-making, execution, and rollback.

Core Features

Manage the three libraries of skills / MCP / plugin

dsh-genome brings the three libraries under registry / metrics / history. Skills can be maintained through auto-absorption; external plugins can be searched and installed via the Store Bridge to absorb into the registry.

Negative feedback servo loop

The loop path is:

  1. Observe: metrics;
  2. Compare: scan;
  3. Decide: select;
  4. Execute: mutate / up / down / rollback;
  5. Re-observe.

The key point here is continuing to observe after execution, rather than just making one change.

Progressive stability

Verified progressive stability mechanisms include threshold gates (dead zones), amplitude limits (maxExpanded), 30s rollback redundancy, stop conditions, and human-machine combination.

Specific thresholds are as follows:

  • Sick = Error rate > 10%
  • Weak = Feedback < 3.0
  • maxExpanded = 6
  • 30s rollback
  • After N consecutive rounds with no improvement, it enters a steady state and stops evolving.

Auto-absorption and metrics

When starting, it scans the following standard roots and automatically writes to registry.skills:

~/.agents/skills
$DSH_HOME/skills

It also reads the skills.list() global layer.

Tools automatically measure by listening to tools/result events. Every tool call writes metrics automatically for Sick/Weak judgment.

Web UI and SKILL commands

After enabling, a “Genome” section appears in the Web UI settings page, and a status bar appears below the composer.

You can read skills/genome/SKILL.md to get the management & evolution methodology, and use the following SKILL commands:

/genome scan|mutate|select|rollback|log|status

Store Bridge

dsh-genome can search the GitHub dsh-plugin topic and install the search results to absorb into the registry.

Verified facts indicate that the Store Bridge search uses the GitHub public API, and the unauthenticated rate limit is 10 times/minute.

Dependencies, profile, and security

dsh-genome only uses Node.js built-in modules: fs, os, path, and built-in fetch, with zero external dependencies.

Other verified points:

  • Profile is configurable; install accepts a profile parameter, default is web;
  • Minimal preset compatibility: shell is not a hard dependency, only install needs it; in minimal mode, plugins are still activated, and install will gracefully degrade with a prompt;
  • Command injection protection: install performs whitelist validation on the spec, rejecting ;, &, |.

Installation and enabling

The installation command listed in the verified facts is provided below. This command targets the web profile:

dsh plugin --profile web add dsh-genome

Verify the configuration after installation:

dsh --profile web --dump-config | grep genome

After restarting, a “Genome” section appears in the Web settings page, and a status bar appears below the composer.

Typical usage

You can operate by following the steps below:

  1. Install and verify:
dsh plugin --profile web add dsh-genome
dsh --profile web --dump-config | grep genome

  1. After restarting, go to the Web settings page and check the “Genome” section and the composer status bar.

  2. Read the methodology file:

skills/genome/SKILL.md
  1. Use SKILL commands:
/genome scan|mutate|select|rollback|log|status
  1. If you need to introduce external plugins, you can search the GitHub dsh-plugin topic via the Store Bridge and then install to absorb into the registry.

Applicable scenarios and notes

Suitable for developers using skills, MCP, and plugins in DSH who want to delegate extension effects to metrics, scan, select, rollback, and stop condition management.

Notes to consider before use:

  • dsh-genome runs with the current dsh process permissions;
  • You should check the source code and license before installing; verified facts show the license is MIT;
  • Store Bridge search relies on the GitHub public API, and the unauthenticated rate limit is 10 times/minute;
  • Store Bridge install performs whitelist validation on the spec, rejecting ;, &, |;
  • In minimal mode, if install requires a shell environment, it will gracefully degrade with a prompt.

Conclusion

The value of dsh-genome lies in putting DSH’s three types of capabilities into the same closed loop that is observable, rollback-able, and stoppable. It does not provide uncontrollable automatic expansion; instead, it observes, compares, decides, executes, and then continues to observe.

Repository address:

  • GitHub: https://github.com/Elohia/dsh-genome

Verified facts do not provide a directory page URL.