What Is dsh-evolve-modes? Composable Agent Workflows for DSH

dsh-evolve-modes is a standalone DeepSeek Harness (DSH) web plugin that makes agent workflows composable, reviewable, and continuously improvable: a compact control next to the input area lets you combine working state, reasoning strategy, quality gates, and self-evolution, ultimately enabling Agent Self Evolving. This guide covers what dsh-evolve-modes is, its core features, the full install/update/uninstall commands, typical combos, and troubleshooting — turning your agent from "one fixed answering style" into "a set of workflow decisions you can recombine for each task".

What Is dsh-evolve-modes?

dsh-evolve-modes solves the problem of rigid, opaque, and hard-to-improve agent behavior by turning working state, reasoning strategy, quality gates, and self-evolution into a set of workflow decisions you can recombine per task. The positioning and facts below all come from the official README (source):

dsh-evolve-modes (GitHub repo graysilver/dsh-evolve-modes) is maintained by graysilver under the MIT license. It does not fork DeepSeek Harness, duplicate the agent loop, or modify core code: after installation, the current combo (e.g. "Normal · Standard · Off · Evolution On") is always shown next to the input area, and a global Self-Evolution Mode settings page manages cross-session learning proposals and approved rules. It is not an exclusive "persona mode" but a set of workflow decisions recombined per task — an out-of-the-box agent workflow plugin in the DSH Plugin ecosystem.

What Are the Core Features of dsh-evolve-modes?

The core capabilities revolve around "compose + review + evolve": a four-dimension combo control, plan mode, first-principles reasoning, two quality gates, and human-reviewed self-evolution learning. These capabilities all come from the official README (source):

• Composable workflow control: the input area shows the current combo, with four adjustable dimensions — working state (normal/plan), reasoning strategy (standard/first-principles), quality gate (off/adversarial review/acceptance review), and self-evolution (off/on).
• Plan mode: delegates to the official DSH plan service, reusing plan persistence and the approval flow instead of reimplementing another plan system.
• First-principles reasoning: writes goals, facts, assumptions, constraints, derivations, and verification into the request header, with Trajectory preserving the evidence.
• Quality gates: adversarial review independently finds risks and reports evidence and gaps; acceptance review checks results against the task and plan, reporting Met, Gap, Unverified, Evidence, and Concrete follow-up.
• Self-evolution mode: defaults to propose mode, triggering an isolated learning batch every 3 parent replies and generating human-reviewable rule proposals; approved rules go into global learned instructions.
• Command support: /evolve-mode commands usable in the web input area or the command API.

How to Install and Enable dsh-evolve-modes?

One pinned-version official command installs it; after restarting the Web profile, the self-evolution control appears next to the input area.

1. Install dsh-evolve-modes: install the pinned version into the DeepSeek Harness Web profile via npm:

dsh plugin --profile web add @graysilver/dsh-evolve-modes@0.3.2

Use this shorthand when the DSH CLI is globally installed; otherwise run npx -y @deepseek-ai/dsh first. Restart the Web profile after installation.

2. Install a pinned Git revision for source auditing: use a fixed commit when auditing or developing:

dsh plugin --profile web add github:GraySilver/dsh-evolve-modes#<trusted-commit>

Git installs execute install-time code, so only install trusted revisions.

3. Enable and confirm it works: after restarting the Web profile, the self-evolution control should appear next to the input-area tools; open the top-level Self-Evolution Mode settings page to manage global learning rules.

4. Update the plugin: re-run the update command when you need a newer version:

dsh plugin --profile web update @graysilver/dsh-evolve-modes

5. Uninstall the plugin: remove it when no longer needed:

dsh plugin --profile web remove @graysilver/dsh-evolve-modes

The command forms above follow the host documentation, and the package name @graysilver/dsh-evolve-modes comes from the README and the plugin entry (source).

Typical dsh-evolve-modes Usage

dsh-evolve-modes works right after installation; the core habit is choosing a combo per task: base combos for daily work, reviews for high-risk answers, and self-evolution for long-term use.

1. Get quick daily work done: choose "Normal · Standard · Off" to keep the execution pace without extra process.

2. Make high-impact decisions: choose "Plan · First-Principles · Off" to research and expose assumptions before entering the DSH plan approval flow:

/evolve-mode working plan
/evolve-mode reasoning first-principles

3. Confident delivery and high-risk answers: pick "Normal · Standard · Acceptance Review" for delivery and "Normal · First-Principles · Adversarial Review" for high-risk answers:

/evolve-mode quality acceptance-review
/evolve-mode quality general-review

4. Accumulate stable personal preferences: enable self-evolution so it generates rule proposals in batches every 3 parent replies by default:

/evolve-mode evolution propose

Proposals are written into global learned instructions only after you apply them manually in the global Self-Evolution Mode settings page (source).

dsh-evolve-modes Troubleshooting

If you hit issues such as "the control does not appear", "version incompatibility", or "no new proposals", start by matching the symptoms below. The troubleshooting points below all come from the official README (source):

1. The self-evolution control does not appear: the symptom is no control next to the input area after enabling. The cause is a Web profile that was not restarted, or a version mismatch with Harness. Fix: restart the Web profile and refresh the browser; confirm the Harness version matches the plugin version.

2. Version incompatibility error: the symptom is a peerDependencies conflict when installing or running. The cause is that plugin 0.3.2 only supports Harness 0.1.1-rc.2. Fix: install plugin 0.3.1 when still on 0.1.0-rc.6, pinning versions instead of relying on floating tags.

3. No self-evolution proposals are generated: the symptom is no new proposals after many replies. The cause is self-evolution being off, or the learning batch not yet reached (one batch per 3 parent replies by default). Fix: enable it with /evolve-mode evolution propose and check the batch size in settings.

4. You want to restore a proposal you ignored by mistake: the symptom is a rule not taking effect or wanting to undo. The cause is that ignoring does not change later behavior. Fix: manually add or edit global rules in the settings page, or restore from the automatic backup created before each change.

Use Cases and Notes

dsh-evolve-modes suits every DSH user who wants controllable, reviewable, and continuously improvable agent behavior, especially plan-driven teams and high-frequency delivery.

Use cases: developers who need different working styles per task; users who want high-risk answers reviewed independently first; and people who want to accumulate long-term personal preferences. In short, the agent improves with use while every change stays reviewable.

Notes: plugin 0.3.2 only supports Harness 0.1.1-rc.2, so pin both plugin and Harness versions in production; self-evolution only proposes by default and applying requires human confirmation; quality reviews add one model call and latency per completed parent reply; learning requests are isolated and do not load AGENTS.md or CLAUDE.md (source).

Project Links

dsh-evolve-modes is an MIT open-source project maintained by graysilver. Plugin detail page: dsh-evolve-modes plugin details.

This page is an independent guide rewritten from the plugin's official README — for the authoritative documentation and the latest changes, defer to the source: graysilver/dsh-evolve-modes. A plugin is third-party code that runs on your machine once installed; inclusion is not an endorsement — review the source before installing.