chenjie1129/deepseek-harness-reliability-governor
chenjie1129★ 2TypeScript最后同步: 2026-08-26
Evidence-gated completion and trusted code verification for DeepSeek Harness agents
README 摘要
DeepSeek Harness Reliability Governor English 简体中文 Unofficial community project. Public beta testers wanted. Try three to five disposable local tasks and report counterexamples through the 15-minute feedback protocol. False certification, false exhaustion, false abstention, repair regression, brittle checks, and Harness compatibility reports are especially useful. An opt-in DeepSeek Harness bundle that lets a user review the proposed evidence contract, then changes completion from a model assertion into a deterministic evidence decision. It does not make an LLM deterministic. It makes a narrower promise: while a reliability contract is active, the agent is steered until observable checks pass, its bounded repair budget is exhausted, or it abstains. Every attempt and terminal outcome is recorded in the durable session log with a content receipt. Evidence status Evidence Current result Claim allowed Keyless Harness AgentLoop fault matrix 9 cases × 10 trials × 2 arms = 180 runs; mechanism gates pass; zero governed false completions and false certifications The active contract and lifecycle enforce declared deterministic checks under scripted faults. Scripted auxiliary-author boundary …
在 GitHub 查看完整 README →分类
🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
★ 69,644
volcengine/OpenVikingSelf-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
★ 34,132
tt-a1i/archifyAgent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
★ 27,048