AI Agent Hub
Back to plugins
dsh-capability-optimizer preview

dsh-capability-optimizer

Web Tools Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install hezhongtang/dsh-capability-optimizer

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install hezhongtang/dsh-capability-optimizer inside DeepSeek Harness to install this plugin; the source is hosted at https://github.com/hezhongtang/dsh-capability-optimizer . Restart dsh web after installation to activate the agent tools.

About this plugin

A single harness has one perspective, yet at a consequential decision, before declaring risky work done, or before significant new code lands, independent evidence from a second model can be genuinely useful. dsh-capability-optimizer turns that consultation into a bounded tool call rather than a copy-paste detour: the agent headlessly invokes the Claude Code CLI under an explicit role contract, and the structured reply it receives is advice to weigh, not instructions to obey.

Core capabilities include built-in advisor, reviewer, and designer roles alongside fully custom contracts, three-level thinking effort (per-call, per-role, global default), one-hop model fallback with metadata, parallel multi-role consultation via consult_panel, and an auto-consult policy that fires lifecycle nudges at file-write and finish anchors, all budgeted per role per session. Security is layered through read-only CLI tools, strict MCP isolation, permission pinning, typed JSON schemas, and bounded output, turns, and wall-clock time.

Ideal for developers and teams working within DeepSeek Harness (web, TUI, or headless) who want a structured, auditable second opinion at critical moments without letting the agent blindly follow an external model output.

Screenshots

Use Cases

  • Consult a reviewer role on a diff before declaring the change complete
  • Ask a designer role for architecture guidance before introducing a major new module
  • Use consult_panel to gather multi-perspective advice in a single parallel call

Best For

  • Developers in DeepSeek Harness (web/TUI/headless) seeking a second opinion at critical moments
  • Teams that want to structure external model consultation as bounded, auditable tool calls
  • Engineering leads with strict requirements for sandboxing and reproducibility