ai-safety

该仓库暂未提供项目说明。

Claude Code-style declarative permission rules for DeepSeek Harness: ordered allow/deny/ask rules with tool-name, argument (glob/regex), and workspace-path matching on the tools/pre-execute waterfall, session-log audit, and HMR reload.

Second-model AI auto-review for DeepSeek Harness approval requests: a read-only reviewer subagent returns structured allow/deny verdicts with reasons, fail-closed by default, fully auditable from the session log (approval/asked -> autoReview/verdict -> approval/decided).

让测试绿了不等于你做对了 — 双轴收敛纪律 (spec-gaming orthogonal axis) 的 DeepSeek Harness 原生实现 | Two-axis convergence discipline for coding agents