@goodandready/dsh-moa
Mixture of Agents (MoA) plugin for DeepSeek Harness with /moa slash command 适合需要使用/moa命令,将提示词分发给多个参考模型融合输出的高级用户。
설치
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:GooDAnDReaDY/dsh-moa⚙️ Configuration (settings.yaml)
Configure presets and model pipelines in settings.yaml or through the Web UI Settings panel (Settings → Plugins → Mixture of Agents):
### Configuration Parameters
| Parameter | Type | Default | Description |
|:---|:---|:---|:---|
| `enabled` | `boolean` | `true` | Master switch for the `/moa` command, turn routing and `POST /dsh-moa/run` (editable in the settings card) |
| `default_preset` | `string` | `"default"` | Preset invoked when typing `/moa
` without an explicit preset |
| `presets` | `array` | `[...]` | Named presets; selected via `/moa
` or `/moa --preset=
` |
| `presets[].reference_models` | `array` | `[...]` | Proposer models queried concurrently during the proposal phase |
| `presets[].aggregator` | `object` | `{...}` | Judge model responsible for synthesis, critique, and winner selection |
| `presets[].ask_clarifying_questions` | `boolean` | `true` | Synthesize a clarifying questionnaire for broad/underspecified prompts (per preset) |
| `presets[].curator_synthesis` | `boolean` | `false` | Curator mode: evaluates strongest parts across candidates using the antipatterns rubric and advises an assembler model |
| `presets[].stream_aggregator` | `boolean` | `true` | Stream judge/aggregator tokens live in real-time with zero TTFT wait |
| `presets[].quorum_enabled` | `boolean` | `false` | Straggler mitigation: proceed with synthesis once >= 60% candidates respond |
| `presets[].grace_period_sec` | `number` | `10` | Grace period in seconds to wait for stragglers after quorum is reached |
| `presets[].aggregator_fallbacks` | `array` | `[]` | Ordered fallback judge models tried if primary aggregator encounters transient errors |
| `presets[].blind_evaluation` | `boolean` | `false` | Anonymize candidate model names for the judge/curator to eliminate family/brand bias |
| `presets[].reference_timeout_sec` | `number` | `60` | Per-candidate execution timeout in seconds |
| `presets[].aggregator_timeout_sec` | `number` | `180` | Aggregator/judge synthesis timeout in seconds |
| `presets[].reference_temperature` / `.aggregator_temperature` | `number` | `0.6` / `0.4` | Sampling temperatures for proposers and judge |
| `presets[].max_tokens` | `number` | `4096` | Max output tokens per model call |
| `presets[].judge_criteria` | `string` | `""` | Optional extra evaluation criteria passed to the judge |
| `prices` | `map` | `{}` | Custom USD-per-1M-token rates (`"provider/model"`, `"provider/*"`, `"*"`) applied to cost estimation |
> **Privacy note:** in refinement mode, readable project files (up to ~16k characters; dotfiles such as `.env*` are excluded) are included in the prompts sent to the configured candidate and judge providers. Avoid running `/moa` in projects whose non-dotfile files contain secrets.
---