dsh-llm-guardian
Provider health, quota, account usage, and local model usage statistics for DeepSeek Harness
206 results
Provider health, quota, account usage, and local model usage statistics for DeepSeek Harness
DSH skill_manage: agent-managed procedural memory. Create/patch/edit/disable/delete skills in ~/.dsh/skills (user scope) or <projectRoot>/.dsh/skills (project scope), with validation, delete guards (agent-created marker, pin, path confinement), disable/enable via disable-model-invocation, and an English trigger-discipline system prompt section. Built on top of skill-filesystem's hot-reload watcher.
Safemode profile enforcer: force-restores ~/.dsh/profiles/safemode to a clean whitelist (dsh-base + dsh-web-app) on start and continuously guards it against modification, so `dsh --profile safemode` always boots with zero third-party plugins. No build scripts: the bundle row does all work.
Timeout guard for ask_user_question: a lost or unanswered question resolves as a structured ASK_TIMEOUT instead of hanging the turn forever
DSH 并发请求监控与门闩:挂钩 llm/stream 瀑布,统计全部在途模型请求并按来源分类(主会话/子代理/插件/压缩/标题)与会话活跃聚合,达到上限 FIFO 排队,防止并发超限被供应商锁号;v1.5.0 新增会话级并发控制——实时给在线活跃会话设置并发数(rootId 解析到顶层会话,含子代理),面板新页签「会话并发」;v1.5.1 异常明细改 tab 切换并新增「按分类×错误信息」汇总报表;v1.5.2 异常明细加「今日/全部」范围切换、逐条明细改倒序、新增历史数据清理(保留最近 N 天 / 分项清理 / dayRetention 自动保留);提供 WebUI 实时面板(含历史清理)、HTTP JSON 端点、状态文件与 concurrency_status / concurrency_session_list / concurrency_session_set 工具。
Source-aware prompt injection protection for DeepSeek Harness
GitHub-only marketplace and guarded lifecycle manager for DeepSeek Harness plugins
DSH WebUI capability controls for skills, MCP servers, tools, prompt injections, approvals, and safety guards at session, project, and global levels.
Audit deepseek-harness (DSH) plugins before install, guard them at runtime. Static verdict + pre-install audit protocol; supply-chain checks (typosquat, OSV); exfiltration & ransomware detection, honeypot canaries, integrity baseline. Alarm-only; blocks confirmed destructive ops.
Reliable todo panel for DSH: survives restarts, verifies completion evidence
Agent-driven kanban board plugin for DeepSeek Harness: one board per workspace, per-task agent sessions, human/agent workflow guards, light/dark UI. Based on the dsh-kanban plugin architecture.
Composable three-tier memory control plane for DeepSeek Harness: persistent runtime context, searchable project documents, pluggable long-term memory, guarded strategies, WebUI, and headless tools.
Independent Codex Guardian-style approval reviewer for DSH.
FileLens — right-column file explorer for DeepSeek Harness: lazy directory tree, multi-format preview (Markdown/JSON/CSV/diff/log/hex/image), search & grep, tabs, inline editing with containment + version-guarded writes.
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.
工具调用规范守卫:对 agent 工具调用参数做字符串匹配,命中危险行为拦截并注入原因(deny),或放行但注入警告(warn),附规则管理面板
DSH plugin guard: on every plugin load checks profile bundle lists for duplicate loader entries (aggregate bundle + sub-plugins listed together → 'duplicate loader entry id' boot crash) and runs site-level health checks — known conflict pairs, vendor @deepseek-ai tree integrity vs lockfile, plugin @deepseek-ai junction integrity — guarding the 2026-08-18 and 2026-08-21 incidents. Read-only diagnostics.
Scan agent skill packages for static security risks. Use for skill audit, security check, prompt-injection review, suspicious scripts, unsafe dependencies, zip package review, and pre-install skill review. Not for runtime monitoring.
A project-agnostic multi-agent collaboration protocol for the DeepSeek Harness: loop guards that stop runaway agent loops, a six-field handoff format, risk-tiered review routing, and a fixed verify-commit-push-review-close sequence.
DSH 规则执行引擎 v3:容器解析 AGENTS.md + 理解器 + 匹配机 + 执行框架
DeepSeek Harness 环境体检插件:磁盘/内存/延迟/缓存膨胀/插件版本落后检测,历史基线趋势对比。只读,不删任何东西。融入了 sys-doctor/pc-guardian 的体检方法论。
Security for DeepSeek Harness in two layers: source-backed pre-install vetting plus fail-closed runtime guardrails and HMAC-chained audit logs. Zero runtime dependencies.
高峰自动会话门:周末模式 + 高峰自动暂停 + 官方源二维判定 + 会话级冻结 + 后端自动重试。
工具调用规范守卫:对 agent 工具调用参数做字符串匹配,命中危险行为拦截并注入原因(deny),或放行但注入警告(warn),附规则管理面板