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).
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.
sofagent is an open-source FDE Harness for AI coding agents — on entry, write your business judgment into files (workflows, ontology, AI nodes); after departure, audit every change against them. 25 codified audit rules, tamper-evident chain, snapshots, and an MCP toolset (104 tools, 11 plugins). Everything can be FDEing.
Make destructive AI-agent actions reversible by default — quarantine + audit + human escalation for rm/git destructive operations. Reliability infrastructure, not a sandbox.
PSA is a multi-classifier behavioral analysis engine for LLM responses. It classifies every sentence into behavioral postures, then derives metrics from the posture sequences to detect adversarial stress, sycophancy, hallucination risk, persuasion techniques, and input pressure — in real time.
Codex-style agent auto-approval for DeepSeek Harness: an independent reviewer model decides allow/deny on the approval answerer chain, fail-closed, with a rationale for every decision (refusals and allows) in a dedicated Approvals tab. · 替我审批：独立复核模型裁决，fail-closed，每次审批留详细理由。
Run an agent unattended for hours and get a morning brief that is an audit, not a summary. Signs a manifest before the work, prices every step at the execution boundary, rolls back a failed verification, and reports what the gate refused. DSH plugin + skill, zero dependencies.
