dsh-sandbox-escalation-fix
Session-aware sandbox escalation compatibility plugin for DeepSeek Harness
13 results
Session-aware sandbox escalation compatibility plugin for DeepSeek Harness
Session replay, fork-tree, audit & compare visualization for DeepSeek Harness — time machine for your agent sessions
Correctover runtime security for DeepSeek Harness (DSH): CCS 7-dimension verification, command-injection/SSRF/credential-exfil blocking, Ed25519 receipts, audit-first. Install with `dsh plugin add dsh-correctover`.
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.
DSH WebUI capability controls for skills, MCP servers, tools, prompt injections, approvals, and safety guards at session, project, and global levels.
Portable DSH WebUI skill security inspection with local rules and optional model review
Audit an agent harness against the harness-evaluation criteria, with machine-enforced evidence validation.
Runtime security guard for DeepSeek Harness (DSH) — blocks command injection, SSRF, credential exfiltration, and destructive operations at runtime.
DSH profile plugin: unconfined (GPU-capable) session sandbox plus per-operation user approval for writes outside the workspace or to protected files.
DSH web plugin: play a notification chime when the agent asks the user a question (ask_user_question / plan review) or requests approval (sandbox escalation, tool approval).
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.
Press Enter to approve the sandbox permission-escalation prompt in the DeepSeek Harness WebUI instead of clicking the Allow once button.
并行多角色代码评审小队:/review 一条命令派出安全/正确性/测试/风格四名只读评审员子代理(每人可单独指定模型与思考强度),汇总为按严重度分组的结构化报告。Parallel multi-role code review squad: /review dispatches security/correctness/tests/style reviewer subagents (each on its own model) and aggregates one severity-grouped report.