What it doesAI
Adds multi-candidate mode to DeepSeek Harness for test-time scaling with a floating whale panel config.
- Floating whale panel configuration
- Auto-runs parallel model candidates
- Verifier selection for candidates
AI-generated from the repo README — for reference only.
Installation
dsh plugin --profile web add github:WintryGrass/dsh-multi-candidateInstall method: GitHub · not yet tested in container (L3+)
Compatibility
| DSH Version | Status |
|---|---|
| not stated | Declared — not tested |
Requirements
- • Node.js: not stated
- • DSH: declared "not stated"
- • External credentials: none detected
Security Report
Automated static scan, not manual review.
No critical or warning findings in the automated scan. The result is valid only for the commit hash shown.
1.1.1dsh manifest6eccdf5bb2a72026-08-206eccdf5bb2a7The scan result is valid for the scanned commit. New commits within 7 days are tolerated (the rating still counts); after 7 days without a rescan the badge shows outdated.
Disclaimer: automated static analysis, not a security guarantee. Always review what you install.
vet verdict: vet: clean · score 100
- R2 · info [likely] — require()（客户端加载器 factory 形参注入，DSH bundle 标准写法）
- R10 · info [heuristic] — 依赖清单：6 项（@deepseek-ai/cordis, @deepseek-ai/dsh-tools, @deepseek-ai/dsh-settings, @deepseek-ai/dsh-system-prompt, @deepseek-ai/schemastery, react，供 LLM 审计供应链；已知漏洞核对见后续 OSV 精确版本查询（网络失败静默降级））
vet is advisory and does not change dsh’s four-tier level.
• Static heuristic scan: done (4 files)
• Dependency vulnerabilities: requires deep audit (L3+)
• Permission sandboxing: requires runtime testing (L4+)
Activity
Last commit 2026-08-20 · activity: Active
• Repo created: 2026-08-20
• Stars: ★ 1 · Forks: 0
• Health: Active — committed within last 30 days