@mokuyoaxis/agent-guard
Harness-neutral reliability infrastructure for AI coding agents: reversible destructive actions, pre-emission redaction, recovery evidence, and a shared Decision Protocol with optional native adapters.
31 results
Harness-neutral reliability infrastructure for AI coding agents: reversible destructive actions, pre-emission redaction, recovery evidence, and a shared Decision Protocol with optional native adapters.
唯稳律(Weiwen's Law)通用因果引擎(白箱呈现)—— DeepSeek Harness (Cordis) 插件实现
Emotional checkpoint & repair for DeepSeek Harness — soothe, fold & seal heated exchanges, then resume from a clean context · 情绪检查点与修复:安抚、折叠封存、干净上下文继续
Static capability disclosure for DeepSeek Harness plugins: capabilities, injected prompts, install scripts, and provenance, each with file:line evidence. Facts, not a safety verdict.
KISS's Law (Weiwen's Law) universal causal engine (white-box presentation). DeepSeek Harness (Cordis) plugin implementation.
DSH permission-gate for the DeepSeek Harness 0.1.5 line: a single self-sufficient, deterministic-first, fail-closed gate covering P0 hard-deny -> P1 session grant -> P2 static rule (allow/deny) chain -> P3 optional LLM semantic classifier -> P4 ask, with command whitelist/blacklist.
LLM auto-review approval answerer for DeepSeek Harness — decides sandbox escalations without a human prompt via a deterministic filter and a clean-context LLM safety review. REQUIRES a patched harness core (see core-patches/)
Runtime security guard for DeepSeek Harness (DSH) — blocks command injection, SSRF, credential exfiltration, and destructive operations at runtime.
Non-intrusive safety belt for DeepSeek Harness: blocks API keys pasted to the agent, blocks secret writes, tracks leak events, and offers a workspace health scan. 非侵入式安全护栏:拦截粘贴给 Agent 的 API 密钥、写入拦截、泄露追踪与工作区安全体检。
A safety harness plugin for DeepSeek Harness (DSH): protected-path enforcement, trash-based safe_delete with undo, last-known-good composition snapshots, pre-restart composition checks, and an audit journal.
Self-protection guardrails for the DeepSeek Harness: protected-path interception, backup-before-destroy, CLI self-recovery commands, and strict sandbox defaulting
Maestro Guard — pre-execute safety gate: native DSH approval prompts, parsed command-surface rules, decision journal
Independent Codex Guardian-style approval reviewer for DSH.
Tiered auto-review for DeepSeek Harness (DSH): static-rule safety net + LLM reviewer + human fallback. Auto-approve safe actions, auto-deny irreversible ones, ask a human for the rest. PURE VIBE CODING - not audited, use at your own risk.
Hard-stop DeepSeek Harness tool use after a per-session call budget.
Fifth DeepSeek Harness mode with configurable Codex, Claude Code, or DSH audits, user-approved remediation turns, progressive Cordis/skill elevation, safety pauses, and Web/TUI review controls.
A dsh policy plugin that gates kubectl writes by kubeconfig context: hard-deny irreversible verbs outside local clusters, ask for the rest.
Declarative Claude Code-style permission rules plus a Codex-style process-level network policy for DeepSeek Harness: ordered allow/deny/ask rules with tool-name, argument (glob/regex), workspace-path, and network-target (domain/ip/port/scheme) matching on
AI Agent runtime authorization & evidence verification — tool-call GuardrailProvider, CCS 7-dimension verification standard, MCP/DSH security scanner, SSRF/command-injection/credential-exfil blocking with Ed25519 signed receipts.
Installing-time safety gate for the DSH plugin market — scans npm / GitHub plugins before the market executes them. · DSH 插件市场安装安全检查:安装前自动检测三方插件风险,安全直接放行,有风险弹窗由你决定。
Guard DeepSeek Harness PowerShell, file, image, and search tools on Windows with workspace, credential, approval, and audit policy.
Auto-reject unanswered permission requests with a model-visible timeout notice
DSH Auto Guard: a Claude Code Auto Mode-like command approval mechanism that adds an LLM safety net on top of full access, with rules, caches, file tracker, and sensitive path gates.
DSH plugin: file-change diff bar with keep/undo summary, and a dangerous-command approval gate with red highlighting for bash/pwsh