dsh-prompt-shield
A runtime prompt-injection shield for DeepSeek Harness tool results.
187 results
A runtime prompt-injection shield for DeepSeek Harness tool results.
对角巷 · 角色面具 — 一键切换AI人格,定制专属对话角色
An integrated starter plugin for newly installed DeepSeek Harness (DSH). Combines whale-girl persona injection, desktop pet, balance widget, notes, reminders, shared long-term memory, and low-token memory tools.
Small DeepSeek Harness memory plugin: the model writes markdown files under ~/.dsh/memory, the host injects them next session. No auto-extract, no vector DB.
Phase 1 Claw identity: inject core workspace files into the agent prompt. 把核心人设文件打进 Claw Agent 提示词。
DSH local cross-session memory plugin: capture turn summaries, inject recent days, and search all historical memory with pure JS keyword matching.
Unofficial community plugin: per-session Engram memory for DeepSeek Harness — spawns an engram MCP child per agent session rooted at the session workspace, registers mem_* tools per agent scope, and injects the Memory Protocol as a system-prompt section.
Per-message session handoff for DeepSeek Harness: a handoff icon next to each message's fork button summarizes the conversation up to that message with an LLM, forks a fresh session carrying the copied history plus the injected summary document, and auto-opens it. The origin session is unchanged.
DSH web extension: live subagent monitor + role-based subagent routing + Claude Code style .dsh/agents/*.md persona injection
Mingleng mcpguard for DeepSeek Harness 鈥?the first security plugin for DSH. Scans skills and MCP configs for prompt injection, homoglyphs, hidden Unicode, dangerous shell and credential leaks.
Post-compaction premise-drift guard plugin for DeepSeek Harness: after a compaction summary drops a critical literal anchor (path, value, error code) it injects a one-shot notice telling the model what it may have lost and how to recover it from the log
Runtime security guard for DeepSeek Harness (DSH) — blocks command injection, SSRF, credential exfiltration, and destructive operations at runtime.
Outbound LLM request header modification for DeepSeek Harness (DSH): client-identity header presets (Codex / Claude Code / opencode), per-request dynamic headers, X-Stainless fingerprint stripping, URL rewriting and Anthropic body patching at the fetch transport layer.
dsh-experience-flywheel · DSH 经验飞轮插件 — agent cross-session experience memory + dual-acceptance gates (auto query/inject via agent/pre-step middleware; exit-1 gate scripts)
System tray support for DSH Desktop: tray icon plus settings to hide the main window on minimize or close, with self-healing boot-time injection and an injection management panel.
Remote access login gate for the DeepSeek Harness Web UI: a password wall in front of non-loopback access, session cookies, an injected browser-side gate, and a Remote Access settings section. The /api RPC surface is authenticated by dsh itself.
DSH LLM response-stream injection/pollution filter: hard-block rare Unicode scripts (Track A) and score-based disposal of control chars / protocol markers / script mixing / spam keywords (Track B) on the llm/stream waterfall.
Context compression for small-context models on the DeepSeek Harness Web GUI: one click condenses the whole conversation into a few sentences, opens a fresh session in the same workspace, and re-injects the summary as the first message so a small model keeps working with a clean, compact context.
Personalization plugin for the DeepSeek Harness (dsh): teach your assistant who you are, who it is, and how it should behave. Adds a Personalization page to dsh settings (custom instructions, nicknames, persona) and injects the content into every agent turn's system prompt.
全局提示注入 / 上下文病毒(context virus)防御插件:tools/pre-execute 扫工具参数、tools/post-execute 扫工具结果、agent/pre-step 扫进入模型前的消息并注入金丝雀守卫、llm/stream 出站消毒;支持 block / quarantine / monitor 三模式、磁盘化可演进病毒库(学习/导入/导出)与审计日志。纯 JS 无原生依赖,profile 层挂载,对所有预设与子代理全局生效。
Attach local files to the conversation context: a composer chip adds files whose content is injected into the model context (agent/pre-step), with a removable ✕ chips strip above the input.
Global and workspace rules for DeepSeek Harness agents: configurable storage paths, system-prompt injection and a rules_read model tool.
Automatic TMCRA recall, injection, and role-separated writeback for DeepSeek Harness
Vision recognition plugin for DeepSeek Harness: paste images into the composer, recognize them via GLM-4V on the host side, and inject the result into the conversation.