16 results
MCP bridge: lets ChatGPT Web create, view, continue and control DeepSeek Harness (DSH) agent sessions through the official MCP protocol. The bridge only connects; DSH keeps its session, agent, tool, approval and workspace security model.
Tuning Engines CLI, MCP server, and Python agent runtime adapters for governed model, agent, skill, and MCP workflows. Fine-tune open-source LLMs, run inference, manage datasets/evaluations, and connect LangGraph or Temporal while Tuning Engines handles policy, audit, usage, and token economics.
DSH web plugin: adds a one-click Compact button and a New Session button to the conversation context meter panel. Compact submits /compact to summarize older history; New Session starts a fresh session in the same workspace (agent preset and permission settings follow the deployment default).
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/)
DSH context-window relief: knowledge base (ctx_index/search), routing enforcement (deny flood tools), sandboxed execution (ctx_execute/batch Think-in-Code), and session continuity (post-compaction restore).
PRO-LONG-style programmatic memory for DeepSeek Harness: appends every session event to a per-workspace log.txt that the agent retrieves with grep/python. Built-in write probe, /prolong status command and permission-deny counter.
X-ray for your DeepSeek Harness — diagnostics for what's actually loaded, why, and what it costs: per-plugin context-tax attribution, per-request token ledger, skill catalog pricing, dependency cascades.
Conversation-local credential handoff that writes through the DSH credential service without returning the secret to model context.
DeepSeek Harness plugin (temporary, pre-upstream-fix): automatically injects a CPython tempfile shim (PYTHONPATH -> sitecustomize) into every confined shell command on the Windows sandbox, so python/pytest tempfile use works with zero extra tools, zero model-context overhead, and zero escalation.
LLM pre-review for sandbox-escalation approvals: an independent-context LLM gate answers sandbox escalation requests before they reach the user, falling back to the user on any failure.
Forgetting engine for AI agents — memory TTL, decay, eviction, audit. The opposite of memory programming.
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。
Static and runtime security guard for the DeepSeek Harness: scans plugins and workspaces for malicious code, context injection and token waste (block/warn/clean), intercepts dangerous runtime tool calls and prompt steps, and exposes /scan, plugin_scan, the scan web panel and a user-managed allowlist. Fully static analysis — never executes scanned code.
A dsh policy plugin that gates kubectl writes by kubeconfig context: hard-deny irreversible verbs outside local clusters, ask for the rest.
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.
DeepSeek Harness plugin: plants a canary string in model-visible context and denies any tool argument or outbound URL that echoes it back