> ⚠️ Archived — merged into dsh-ops-kit > This repository shipped one of five skills that used to be published as five separate packages. They are now maintained together as a single bundle so nobody has to install — and the ecosystem does not have to index — the same skill five times. Please switch to dsh-ops-kit. > > 已归档 — 已合并至 dsh-ops-kit > 本仓库此前是五个独立发布包之一,其功能现已与另外四个包一起合并维护为单一 bundle,避免同一批 skill 被拆成五份重复发布、重复索引。请改用 dsh-ops-kit。
---
dsh-benchmark-evidence
Benchmark evidence and integrity gates for DeepSeek Harness. It generalizes the run-manifest, fail-closed preflight, artifact inventory, result-integrity, cleanup, and error-bucketing practices used around Harbor and Terminal-Bench.
Tools:
dsh_benchmark_manifest: create a frozen identity manifest without starting a run.dsh_benchmark_gate: distinguish a clean score candidate from incomplete or contaminated results.
npm install dsh-benchmark-evidence
# or: pnpm add dsh-benchmark-evidenceCompleted is not treated as scored automatically. The package never starts a benchmark, handles credentials, uploads artifacts, or mutates a run directory.
npm run build && npm run typecheck && npm test && npm pack --dry-run