This article is derived from the plugin's already-indexed fields (wiki / faq / compatibility_json / readme), not freshly generated by AI. Source field is noted at the end of each section.

Quick start

dsh-evolve-modes

— source: plugin_wiki.wiki_content

Install & verify

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

Run the command above in your DSH Web Profile. Then enable the plugin in the plugin list.

— source: plugins.install

Key points

• 每个源会话最多取最近 100 条学习消息。
• 保留每轮完整的用户消息，以及该轮最后一个可见的助手消息。
• 助手消息只作为上下文；超过 2000 个字符时保留开头 1000 个字符和结尾 1000 个字符，中间以 ... 代替。
• 提议的证据必须逐字来自用户消息。助手的推断、一次性任务细节、实现结果、沉默或“没有再次提到”都不能单独成为规则证据。
• 不继承父 Agent 的工作型上下文；

— source: plugin_wiki.readme_en (fallback readme_raw)

FAQ

Where can I find the plugin controls after installation?

After installing and restarting the Web profile, the self-evolution mode controls will appear on the left side of the input area (conversation.input.left slot), and the self-evolution global settings entry is located in the Self-evolution mode section of the top-level Settings page.

Will the plugin modify DeepSeek Harness core code?

No. cordis.patch.yml only inserts the plugin's own injection description and does not fork the core code; the plugin state is saved in its own storage domain (graysilver_dsh_evolve_modes / graysilver_dsh_evolve_modes_evolution), and does not modify AGENTS.md or CLAUDE.md.

How many additional model calls will adversarial review/acceptance review add?

Each time the parent Agent completes a response, the quality gate will additionally trigger a subagent (fork subagent) call and generate a Markdown review report. Self-evolution analysis triggers an isolated LLM call after each learning batch (default 3 parent responses) is completed.

Do the self-evolution mode rules take effect automatically?

No. Learning results only generate proposals pending review. They only become approved rules after manually clicking Apply in the Self-evolution mode settings page, and are injected into subsequent system prompts through the tagged segment.

What context will the learning request read from the parent session?

Learning requests are isolated: they use an independent persona and system prompt, do not inherit the parent session history, do not inherit the parent Agent's working context, do not carry tools, do not create learning subagents, and do not load the source workspace's AGENTS.md/CLAUDE.md.

Where is the data stored? How to uninstall?

Session-level state is stored in the graysilver_dsh_evolve_modes storage domain, and cross-session self-evolution rules/proposals/backups/learning records are stored in the graysilver_dsh_evolve_modes_evolution domain; after uninstalling via dsh plugin remove, these data will remain in the host storage and can continue to be reused upon reinstallation.

What is the relationship with DSH's built-in plan mode?

The plugin directly reuses the official @deepseek-ai/dsh-plan-mode service (planMode service) and exit_plan_mode approval process, and on this basis uses tools/pre-execute to restrict the tool whitelist available in plan mode.

Will session records from old versions be automatically migrated?

Yes. 0.3.x will copy data from the old storage domains graysilver_task_modes and graysilver_task_modes_evolution, and map the old single-mode aliases (normal/first-principles/adversarial-review) to new fields, and will not overwrite existing data in the new domain.

— source: plugin_wiki.faq_json

Compatibility

• DSH: 0.1.0-rc.6+
• Node: >=22.19.0

— source: plugin_wiki.compatibility_json

Pitfalls

Review the upstream repo before installing. This guide is auto-derived from indexed fields and may lag the latest release. If anything contradicts the official docs, treat the upstream source as authoritative.

— source: general rule