SEMANTIC DIRECTORY · plugin-directory-semantic-v3-18

Model & Agent Quality Evaluation

以受控实验和统计证据评估模型、提示词、RAG 或智能体配置,包括基准测试、A/B、回归、盲测、漂移检测、连通性诊断和发布治理;不承担日常运行遥测。

Use keyword search
Source catalog discovery.composite.2026-09-06.v1-0-5.b92b985e31dfCanonical structure · 823 nodes · 14829 plugins

Leaf membership

22 plugins

benchMauricioPerera/benchCatalog AnalyzedDSH plugin from MauricioPerera/benchTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals0 starsdeepseekharnessevalsuyoumo/deepseekharnessevalCatalog AnalyzedDSH plugin from suyoumo/deepseekharnessevalTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals5 starsdeepsiper-enthea8b-is/deepsiper-entheaCatalog AnalyzedSovereign, agent-driven LLM evaluation harness forked from deepseek-harness (dsh) — Cordis plugin architecture, EntheAI backends, and multi-model benchmarkingTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals1 starsdsh-agent-evolutionFriendsHL/dsh-agent-evolutionCatalog AnalyzedPreset-composed Agent experiments and evolution primitives for DeepSeek HarnessTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals0 starsdsh-agent-evolvejonah791/dsh-agent-evolveCatalog AnalyzedDSH self-built pluginUI EnhancementsGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals1 starsdsh-bench1105623876/dsh-benchCatalog Analyzed一个用于做模型对照实验的DeepSeek HarnessPluginTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals1 starsdsh-blind-arenachanger-changer/dsh-blind-arenaCatalog AnalyzedA blind, fair, local DSH Web arena: same task, isolated worktrees, shared verification, judge before reveal.Tools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals2 starsdsh-design-qadsh-design-qaCatalog AnalyzedDesign-fidelity QA for text-only models: a `deepseek_vision` tool borrows an eye from any OpenAI-compatible vision route, so the model can judge whether an implementation matches its mock — shipped with the benchmark behind that judgement (four fixtures, 23 injected defects, raw transcripts) and the questioning discipline it depends on.Vision & MultimodalnpmGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency audit0 reported in bounded auditArtifact0.1.4Evidence updated2026-09-06Public signals23 stars · 873 downloadsdsh-gray-testLiLiLi-Zi/dsh-gray-testCatalog Analyzeddsh gray-model detection plugin — launch N concurrent sessions, stream reasoning chains, auto-classify gray models (I'm+ing > Let me×5), normal models (Let me≥3), and abnormal models (no features). Supports standard/PTC/minimal/cordis presets.Sessions & MessagesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals1 starsdsh-harbor-evolutiondsh-harbor-evolutionCatalog AnalyzedDeepSeek Harness plugin and Skill for Harbor Candidate and Historical Session evaluation workflows.Sessions & MessagesnpmGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency audit0 reported in bounded auditArtifact0.9.5Evidence updated2026-09-08Public signalsNo public usage signalsdsh-model-deploylhwwxy/dsh-model-deployCatalog AnalyzedLLM model selection & deployment analysis tool for DeepSeek Harness: deployability, VRAM, TTFT, latency, throughput and power for 38 models × 20 GPUs/NPUsTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals0 starsdsh-models-radardsh-models-radarCatalog AnalyzedDisplays CodexRadar model capability benchmarks in a Settings page and shows the selected session model's DeepSWE score beside the composer.Models & ProvidersnpmGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency audit0 reported in bounded auditArtifact0.1.21Evidence updated2026-09-07Public signals2 stars · 869 downloadsdsh-provider-probeWanbinyu/dsh-provider-probeCatalog AnalyzedManual provider connectivity and latency checks for DeepSeek HarnessTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals2 starsdsh-purgeYuJunZhiXue/dsh-purgeCatalog AnalyzedDeepSeek Harness 破甲:让所有模型都能破甲,不同模型可换不同提示词;默认提示词面向国模「小码酱」。Jailbreak for every model — swap prompts per model. 求 Star 收藏 ⭐Tools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals389 starsdsh-research-labHarzva/dsh-research-labCatalog AnalyzedDSH-native multi-runtime baseline, ablation, and reproducible evaluation control planeTools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals0 starsdsh-self-evolutionLhy723/dsh-self-evolutionCatalog AnalyzedBenchmark-driven self-evolution for DeepSeek Harness · 冻结基准上的 Agent Profile 自我进化:评测 → Candidate → 严格接受/回滚Tools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals3 starsdsh-self-harness-toolsSuTang-vain/dsh-self-harness-toolsCatalog AnalyzedTwo self-contained DeepSeek Harness plugins: dsh-prior-probe (frozen probe-battery tools, official bundle) and dsh-evidence-dashboard (archive dashboard).Tools & CapabilitiesGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals1 starsdsh-skill-evalrenjianguojinqianfan/dsh-skill-evalCatalog AnalyzedDSH Plugin:用 LLM judge 评测技能 description 的触发准确率(欠触发/过触发)SkillsGitHubCompatibilityDSH 0.1.2-rc.1 · webDependency auditNot testedArtifactUnresolvedEvidence updated2026-09-10Public signals2 stars

Categories are public semantic index facts, not a final recommendation. Compatibility, permissions, and source evidence remain separate.