verification

Privacy-minimal heuristic per-turn verification summaries for DeepSeek Harness

Keeps a coding agent's green claims honest: verification runs are recorded unmasked, and done is blocked when the evidence is stale, failed or masked.

变更证明（ChangeProof）— DeepSeek Harness 插件：代码改动后确认改动的行真的被测试覆盖到

The quality gate for agent-built web apps. Real-browser acceptance checks — CLI, MCP (Claude Code / Cursor / Copilot), GitHub Action. Agents self-test and pass; real browsers tell the truth.

Cordis-native, evidence-driven full-stack engineering discipline for DeepSeek Harness agents

Acceptance-driven autonomous completion for DeepSeek Harness. Done means verified.

Verification toolkit for DeepSeek Harness agents: evidence-based claim checking with line citations, config validation (JSON/YAML), HTTP URL, npm and GitHub submission-readiness probes.

该仓库暂未提供项目说明。

VAP - Verifiable Agent Protocol: a tested trust machine for agent-to-agent messages (zero third-party dependencies, 217 tests, full experiment ledger)

LLM-as-a-Verifier plugin for DeepSeek Harness — fine-grained reward tools (verify_compare / verify_select / verify_track) with Probabilistic Pivot Tournament, plus a Best-of-N conversation mode with a Web settings panel.

Three-plane probabilistic verifier for DeepSeek Harness: tiered logprob/sample/template judges, seeded tournament best-of-N, layered context reflux.

AI 说"做完了"？先让它拿证据：目标盖章前强制跑测试，收工前再查一遍 — DeepSeek Harness 插件 / Make your dsh agent prove it's done: gated goal completion + turn-end verification