dsh-api-relay-audit
DeepSeek Harness bundle for running API Relay Audit locally
51 results
DeepSeek Harness bundle for running API Relay Audit locally
Local-first, self-hosted AI agent runtime with Claude Managed Agents-style APIs, sandboxed sessions, memory, tools, audit, replay, and a local Console.
DSH redteam security research modes and runtime plugins, managed from one settings page: deploy the nine presets, install, update and uninstall the seventeen runtime plugins for the current profile.
Sandbox-first automatic permission policy for DeepSeek Harness
Read-only security & compliance toolkit for DeepSeek Harness: prompt-injection detection (rule engine with a pluggable model classifier), Chinese-PII and structured-JSON redaction, and a local configuration security audit that emits redacted, reproducible, self-checksummed risk reports.
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.
侧边聊天 Plus (Side chat Plus) — DSH web plugin: select text in a conversation and ask it in a per-session side chat (hidden ordinary session, own right-side panel, inherits model/effort/permissions, lookup toggle).
Model-based permission approval (approve-for-me) for DeepSeek Harness: an approval/request answerer backed by a separate reviewer model
Model-reviewed Auto Approve permission preset for the DeepSeek Harness Web UI
Enhanced ACP server for DeepSeek Harness: block-level streaming, usage/stat telemetry (cache hit rate, token speed, input/output tokens, context length, turns, tool timing), model & reasoning-effort switching, and permission-preset control over the ACP wire (Zed-friendly)
Fully customizable keyboard shortcuts for the DeepSeek Harness WebUI — macOS-first defaults, 34 pre-registered features across session/view/clipboard/model/permission/system groups, one-click recording to add your own bindings, silent permission cycling with color feedback.
Fail-closed microsandbox microVM provider and model-facing guest tools
CC-style auto mode for DeepSeek Harness: deterministic deny/allow rules + pre-execute gate + model-agnostic two-stage classifier. Two-state (allow/reject) classifier since 0.8.0. TypeScript rewrite merging dsh-auto-mode v0.4.1 with Nuo-cl/dsh-auto-mode native integration.
QuickJS/WASM-isolated WorkflowEngine for running model-written DeepSeek Harness orchestration with bounded resource controls.
DSH tool-call compatibility: strips sandbox_permissions/justification from tool-call arguments for third-party models (GPT etc.), and lets the user skip possibly-stuck tool calls with an LLM-facing notice.
Model-facing typed tools wrapping the frozen, reviewed underseal adapter for the DeepSeek Harness
DeepSeek Harness plugin: auto-detect secrets pasted into the composer, store them in the official credentials seam, and send [secret:REF] placeholders to the model instead of the value.
Agent-decided approvals for DeepSeek Harness: a workspace-write base permission mode where an independent approval subagent judges every sandbox escalation (risky operations are rejected), with a configurable approval model and a per-session audit trail in the conversation window's 审批 tab.
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.
Decision-driven AI company orchestration with HR governance, monetary budgets, model discovery, and deterministic audit for DeepSeek Harness
Portable DSH WebUI skill security inspection with local rules and optional model review
Transport-layer data masking for deepseek-harness (dsh): sensitive values never leave the process — the model sees placeholders, you see real values restored live in the stream.
Better config hub for DeepSeek Harness: one page showing sandbox mode, default model, workspaces, settings namespaces and copy-ready composition templates.
Strip sandbox_permissions / justification from model tool-call arguments so sandbox escalation is never triggered when the session already has sufficient permission.