dsh-approval-core：DSH 自动审批决策管道（fail-safe 加固版）
dsh-approval-core 是 lyaoliu 维护的 DSH admin-security 插件，fork 自 dsh-approval-gate 的加固版。工作区内写文件沙箱直接放行；越界 neutral 类在 riskyThreshold 与学习指纹满足后可 auto-learned 放行；硬风险类别、13 条正则危险清单与 denyKeywords、分类器失败均 fail-safe 转人工。v0.2.0 提供 allowlist.json 热读、受限 POST /api/auto-approve/rules、四级配置 UI，以及审批 tab 的 diff 与块级撤销。安装命令见 README：dsh plugin --profile web add ./dsh-approval-core。MIT 许可。
Read Moredsh-auto-pass: A Model Review-Driven Automated Approval Plugin
The `dsh-auto-pass` plugin introduces an intelligent automatic approval mechanism to DeepSeek Harness (DSH), aiming to resolve the conflict between the low efficiency of manual approval and security risks. Based on a single-pass model review mechanism, it combines blacklist/whitelist management with a denial-list toggle to achieve automated processing while ensuring security. By configuring thresholds, the system can automatically learn and generate a project-level whitelist, significantly improving operational efficiency and supporting a visual approval-record panel. This plugin takes effect only when the “Auto Approval” permission preset is used.
Read MoreDeepSeek Harness Automatic Approval and Review Plugin (dsh-auto-review)
`dsh-auto-review` is an automatic review plugin (v0.4.2) for DeepSeek Harness, designed to address pain points in tool invocation and permission management. Through an independent review layer, it distinguishes server-side failures (such as timeouts and errors) from dangerous operations, defaulting to manual approval rather than rejection. The plugin provides a 90-second total deadline, a complete review history, and fine-grained rule management (such as filtering by session, tool, and parameter). After installation, users can select “Approve for Me” in a session, leveraging Codex-style mechanisms to keep operations controllable and secure.
Read MoreDeepSeek Harness Plugin: dsh-approval-assistant (Approval Assistant)
The DeepSeek Harness plugin dsh, maintained by 0neKbyte, is designed to address issues in AI agent development such as delayed approval workflows, low automation, and lack of feedback. It provides end-to-end approval assistance, with core features including approval and completion reminders (supporting pop-ups, notifications, and sounds), automated approvals (for all requests or based on keyword rules), and model query reminders. Users can install it via official commands and independently configure each reminder channel in the settings. When using it, note the required browser permissions and environment requirements.
Read Moredsh-auto-approval: A Guardian-Style Auto-Approval Plugin for DeepSeek Harness
The `dsh-auto-approval` plugin adds a Guardian mode to DeepSeek Harness (DSH) by using an LLM subagent to assess risk and validate local state before operations. Its core features include session analysis, local filesystem validation, and a four-field adjudication mechanism designed to automatically allow low-risk actions and only prompt for confirmation when the outcome is uncertain. After installation, users must restart DSH and enable the feature via the interface or configuration. It supports a fail-closed security design. When using it, note that the model must support Function Calling, and because the adjudication tool can read absolute paths, it should not be enabled on shared machines.
Read Moredsh-approve-for-me: Auto-Approval Mode Plugin for DeepSeek Harness
`shifan3/dsh-approve-for-me` is a host plugin for DeepSeek Harness (DSH) designed to enhance the security of approval mode. It leverages a large language model (LLM) to automatically evaluate and execute routine approval requests based on conversation summaries, intercept high-risk destructive commands (e.g., `rm -rf`, `drop table`), and support audit logging. Users can flexibly configure the LLM provider, timeout duration, and summary length via a configuration file, making it suitable for developers seeking a balance between automated approvals and manual security checks.
Read Moredsh-guardian-approval: Independent Model-Driven Automated Approval Plugin
The DSH Auto-Approval Plugin provides Codex Guardian–style safety protections for agents through an independent review model. Core features include an independent review channel and configuration, strict policies based on a risk × authorization matrix (accepting only explicit user instructions as authorization; file content is not treated as a basis for authorization), a tri-state circuit breaker to prevent abuse, sidecar audit logging, and a fail-closed mechanism. The plugin supports pre-reading outgoing file content for review and is compatible with multiple API styles, aiming to decouple review logic from the agent model and ensure the safety of agent operations.
Read Moredsh-auto-review: Automatic Approval Decision Plugin for DeepSeek Harness
`dsh-auto-review` is an MIT-licensed plugin for DeepSeek Harness, specifically designed to address the sandbox upgrade approval issue for actions exceeding `workspace-write` permissions when the `approval policy` is set to `'auto'`. It acts as an answerer for `approval/request` and utilizes a dual mechanism of "deterministic filtering + LLM review" to achieve automatic decision-making without human intervention. Core capabilities include: first, performing deterministic filtering to immediately reject or allow clear commands via `hardline` and regex rules; gray-area requests then undergo a clean-context LLM security review, making judgments based solely on the tool name, reason, and command to avoid carrying full session context. The plugin supports...
Read MoreDSH Plugin dsh-auto-approval-plugin: Insert an automatic approval layer between Workspace Write and Full access
Developed by StyxNether, dsh-auto-approval-plugin is a permission approval plugin for DeepSeek Harness (DSH) designed to bridge the gap between Workspace Write and Full access permission levels. The plugin introduces an auto-approval preset and registers listeners along the approval path to perform automatic determination based on actual tool parameters in the session log, rather than model-generated justifications. If the target of the request falls within the user-configured trusted zone (which supports realpath resolution to prevent symbolic link bypass) or matches a harmless pattern, the plugin will automatically approve that specific call (allowed-once); otherwise, the request will be forwarded to human confirmation, and the plugin itself will not execute.
Read MoreDSH-Approval-Gate: DeepSeek Harness Automatic Approval Gating Plugin
dsh-approval-gate is an MIT-licensed plugin for DeepSeek Harness, designed to balance the efficiency of automated approval with security. It leverages Flash models to perform risk pre-assessment on sandbox escape operations, categorizing them into SAFE and RISKY types. For reversible, low-risk operations, the plugin uses a confirmation-based learning mechanism to automatically approve them, reducing repetitive manual confirmations. For the five categories of hard-risk operations—deletions, credential handling, remote/production environments, system paths, and irreversible batch operations—it always routes to manual confirmation to ensure critical security. The plugin provides a manual review UI to display the approval timeline and supports hot-reloading of configurations. Version v0.5.0 introduces file change comparison and one-click revert functionality, allowing users to view unified diffs and restore files based on session snapshots. Users need to install it via the command line.
Read Moredsh-auto-approve: Adding Auto Permission Level Between Workspace Write and Full Access
dsh-auto-approve inserts an auto permission tier between workspace-write and danger-full-access for DeepSeek Harness. The sandbox boundary is identical to workspace-write, with routine sandbox escalations approved in one go by a classification model; cases hitting a deterministic danger list, where the classifier is uncertain, or due to timeout or malformed responses are still escalated to manual approval. The plugin supports Web, TUI, and Desktop frontends and can be installed via dsh plugin and enabled by activating the Auto tier in Permissions.
Read More