maiziman/deepseek-harness-portable--plugins-dsh-model-capabilities1

@maiziman/dsh-model-capabilities

Automatic reasoning and image-input capability discovery for custom DeepSeek Harness model providers

包名
@maiziman/dsh-model-capabilities
版本
0.1.0
许可证
MIT
最近更新
2026年8月30日

安装

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:maiziman/deepseek-harness-portable#b8ebad37978590c0c55ea2f041a10338b1a33b8e&path:plugins/dsh-model-capabilities

Configuration

The shipped Bundle uses:

- insert:
    - id: model-capabilities
      name: '@maiziman/dsh-model-capabilities'
      config:
        metadataDiscovery: true
        activeProbePolicy: local-only
        reasoningProbe: true
        visionProbe: true
        reasoningProbeEfforts: [low, high, max]
        probeConcurrency: 1
        requestTimeoutMs: 10000
        probeMaxTokens: 128
FieldMeaning
metadataDiscoveryRead non-generating model metadata.
activeProbePolicynever, local-only, or always; local-only recognizes exact localhost plus literal loopback, link-local, RFC 1918, and unique-local IPv6 addresses. DNS names require always.
reasoningProbeAllow reasoning requests when the active-probe policy permits them.
visionProbeAllow the inline image request when the active-probe policy permits it.
reasoningProbeEffortsEffort spellings tested independently; only positively observed levels are saved.
probeConcurrencyMaximum simultaneous reasoning probes for one model.
requestTimeoutMsBound for each metadata or active request.
probeMaxTokensMaximum output tokens requested by a probe.

To override the defaults, target the model-capabilities row in the profile's own cordis.patch.yml. A later row configuration replaces the complete earlier configuration, so restate every field you want to keep.