What is this?
This is dsh-evolve-modes, an independent Web plugin for DeepSeek Harness (DSH) that adds a compact set of workflow controls — work state, thinking strategy, quality gate, and self-evolution — so you can compose how an agent works on each task, with self-improvement reviewed by a human before anything is applied.
This is dsh-evolve-modes, an independent Web plugin for DeepSeek Harness (DSH) that adds a compact set of workflow controls — work state, thinking strategy, quality gate, and self-evolution — so you can compose how an agent works on each task, with self-improvement reviewed by a human before anything is applied.Does this modify DeepSeek Harness itself, or run alongside it?It's a standalone Web plugin: it doesn't fork DeepSeek Harness, doesn't copy its Agent loop, and doesn't touch core code. It only adds a control widget in the input area and its own global settings page.
It's a standalone Web plugin: it doesn't fork DeepSeek Harness, doesn't copy its Agent loop, and doesn't touch core code. It only adds a control widget in the input area and its own global settings page.Which DeepSeek Harness version do I need?Plugin version 0.4.0 requires DSH 0.1.1-rc.2. If you're still on Harness 0.1.0-rc.6, install plugin version 0.3.1 instead.
Plugin version 0.4.0 requires DSH 0.1.1-rc.2. If you're still on Harness 0.1.0-rc.6, install plugin version 0.3.1 instead.Does turning on self-evolution automatically change my agent's behavior?No — self-evolution defaults to 'Propose' mode: it only analyzes sessions in batches (every 3 completed replies by default) and generates candidate rules for human review; nothing is applied automatically, and AGENTS.md is never modified.
No — self-evolution defaults to 'Propose' mode: it only analyzes sessions in batches (every 3 completed replies by default) and generates candidate rules for human review; nothing is applied automatically, and AGENTS.md is never modified.What are the four dimensions you combine per task?Work state (normal or plan), thinking strategy (standard, first-principles, or Grilling), quality gate (off, adversarial review, or acceptance review), and self-evolution (off or on). They're independent settings you recombine per task, not fixed persona modes.
Work state (normal or plan), thinking strategy (standard, first-principles, or Grilling), quality gate (off, adversarial review, or acceptance review), and self-evolution (off or on). They're independent settings you recombine per task, not fixed persona modes.What does the Grilling thinking strategy do?It maps out decisions and their dependencies, asks one round of numbered questions with recommended answers at a time, investigates facts it can discover on its own, and waits for your confirmation before implementing anything.
It maps out decisions and their dependencies, asks one round of numbered questions with recommended answers at a time, investigates facts it can discover on its own, and waits for your confirmation before implementing anything.What's the difference between adversarial review and acceptance review?Adversarial review runs an independent agent that hunts for unmet requirements, unsupported claims, gaps, regressions and security risks on its own; acceptance review instead checks the finished result against the task and any approved plan, without silently rewriting or retrying the answer.
Adversarial review runs an independent agent that hunts for unmet requirements, unsupported claims, gaps, regressions and security risks on its own; acceptance review instead checks the finished result against the task and any approved plan, without silently rewriting or retrying the answer.Can I still use the old single-mode settings like 'adversarial-review'?Yes — legacy aliases such as normal, first-principles, and adversarial-review still work; they get mapped onto the new work-state/thinking/quality-gate combination while your current self-evolution setting is preserved.
Yes — legacy aliases such as normal, first-principles, and adversarial-review still work; they get mapped onto the new work-state/thinking/quality-gate combination while your current self-evolution setting is preserved.What are similar projects?deepseek-harness, dsh-router-standard, Proteus, cocode, dsh-commandcode-provider, dsh-plugin-bridge, deepseek-harness-desktop, dsh-agent-team-gui
deepseek-harness, dsh-router-standard, Proteus, cocode, dsh-commandcode-provider, dsh-plugin-bridge, deepseek-harness-desktop, dsh-agent-team-guiWhich topics is it in?self-evolving-agents, agent-harness, dsh, harness, dsh-plugin, deepseek-harness, prompt-engineering, plugin, agent, npm, chinese, ai-agents
self-evolving-agents, agent-harness, dsh, harness, dsh-plugin, deepseek-harness, prompt-engineering, plugin, agent, npm, chinese, ai-agents