laodonge/col-dsh-plugin: Context Organization Layer for DeepSeek Harness
laodonge/col-dsh-plugin is a plugin designed for DeepSeek Harness (DSH), intended to address the issues of tight coupling between context management and execution logic and complex permission control. It introduces persistent context slots and provides core features such as fail-closed permission checks, two-stage auditing, permission pruning, and zero runtime dependencies. Developers should be aware of the security risk of running with the current process's permissions and are advised to review the source code before installation.
Read Moredsh-risk-gate: A DeepSeek Harness Plugin for Semantic Auto-Grading and Progressive Authorization
The `dsh-risk-gate` plugin in DeepSeek Harness provides a local semantic analysis–based mechanism for operation tiers and progressive authorization. It classifies tool calls into safe, risky, and redline levels, then automatically enforces allow, ask, or deny actions without relying on external LLMs. It supports regular-expression checks for Bash commands, enables quiet automatic pass-through via configurable thresholds, and can override behavior for specific tools. This solution is well suited to quickly building a lightweight security control layer within the DSH ecosystem.
Read Moredsh-write-protect: Read-Only Protection for Workspace Paths
The DeepSeek Harness (DSH) plugin `dsh-write-protect` addresses the limitation of the original sandbox, which only supported “all write” or “all read” modes, by allowing developers to apply read-only protection to specific paths (such as `.git`) in `workspace-write` mode. Maintained by azazo1, the plugin supports configuration-based interception of tool writes and command sandbox access, and provides session-level temporary authorization. It also supports placing `.readonly` rule files and configuring additional writable root directories. Note that the plugin requires a DSH engine version of 0.2.0-rc.1 or higher and has compatibility issues on Windows.
Read Moredsh-access-mode: DeepSeek Harness Session Access Mode Plugin
`dsh-access-mode` is a plugin for DeepSeek Harness (DSH) designed to address the limitations of native permission preset balance and the high cost of modifying core code. By providing three session-level access modes—Default / No Edit / Auto—it enables users to finely control the AI’s tool invocations, sandbox policies, and approval policies without modifying the core code of Harness. **Core features**: non-intrusive integration, reuse of native services, and session persistence support. **Mode descriptions**: Default uses a conservative `workspace-write` boundary; No Edit removes file restrictions while only restricting editing; Auto fully permits access. **Security note**: No Edit and
Read Moredsh-agent-gate: Permission Gate · Tool Interception by Tiered Permissions
dsh-agent-gate is a permission gate plugin for DeepSeek Harness, designed to provide fine-grained control over Agent tool invocations and resource access through layered interception and approval mechanisms. Built on the `tools.pre-execute` and `tools.guard` hooks, it supports skill/MCP filtering, policy reuse, and audit logging. Once uninstalled, the interception mechanism stops, but the official file sandbox permissions continue to be independently maintained by the permissions plugin. It can be installed via the `dsh plugin` command. Note that it only intercepts Claw sessions, and it is recommended to use it alongside `dsh-session-permissions`.
Read Moredsh-agent-delegate: Delegation Governance and Workflow Management for DeepSeek Harness
`dsh-agent-delegate` is a delegated-agent plugin designed to resolve issues in DSH sub-agent invocation—such as uncontrolled depth, role overreach, and concurrency overflow—through code-level controls. It provides depth and concurrency control, role filtering, write isolation via independent Git worktrees, and a patch-based result-merge mechanism. The plugin supports configuration-driven limits on delegation depth, sub-agent count, and permission propagation, ensuring that sub-agent changes are merged only after acceptance by the parent task. It is well suited to scenarios requiring strict control over sub-agent behavior.
Read Moredsh-plan-preset: Map DSH Read-Only to Plan Mode
`dsh-plan-preset` is a lightweight configuration plugin for DeepSeek Harness (DSH). It leverages the Cordis Patch mechanism to map the DSH Web UI's "read-only" permission to an OpenCode-style "Plan" mode. Once installed, this mode enforces strict semantic isolation: file operations are restricted to read-only access, web search is available, and any privilege escalation requests are prohibited. This addresses the differing permission requirements between an agent's "research/planning" and "execution" phases, blocks out-of-scope operations, and provides teams with a standardized workflow permission management solution.
Read Moredsh-session-permissions: Session-level Permission Control
`dsh-session-permissions` is a plugin for DeepSeek Harness (DSH) that provides session-level permission control, designed to address security risks caused by overlapping permissions across multiple layers. It determines effective permissions by calculating the intersection of official permissions, Agent policies, and local session overrides. The plugin offers a sandbox pinning mechanism for Claw sessions, ensuring that permissions can only be tightened, not relaxed, and supports fine-grained control over tool and file access. A service restart is required after installation, while the pinning mechanism remains in effect for sessions that are already active after uninstallation.
Read Moredsh-workspace-scope-selection: The Fourth Permission Option for DeepSeek Harness Sessions
The DeepSeek Harness plugin `dsh-workspace-scope-selection` provides a “Selected Workspace Write” option for session permissions, allowing users to precisely specify writable directories via a directory tree editor and balancing security with flexibility. The plugin enforces this restriction in filesystem and Shell tools and supports session persistence. Note that on Windows, only writes to the workspace and temporary areas are supported; out-of-scope write operations require approval escalation.
Read MoreDeepSeek Harness Permissions Rule Engine: dsh-permissions
dsh-permissions is a rule engine plugin for DeepSeek Harness that enables fine-grained security control. It supports four priority levels, layered global and workspace scopes, and rule management through a visual editor. Core features include wildcard matching for file paths, interception by command prefix, and general permission hard-deny mechanisms. All rules are persisted and injected into system prompts, with built-in safe defaults protecting sensitive directories and commands. The plugin is designed to narrow the sandbox scope and provide low-cost, fine-grained policy management.
Read Moredsh-permgate: Adding Fine-Grained Permission Control to DSH
dsh-permgate is an MIT-licensed open-source plugin designed for DeepSeek Harness (DSH), aimed at refining permission control for AI tool invocations from coarse-grained session-level authorization down to the tool-call level. The plugin supports independent review by categories such as directory access, command execution, file read/write, subagent startup, and repetitive operations. Its core features include global and project dual-level configuration, whitelist/blacklist exception management, shortcut tool default value settings, custom rules, approval pop-ups (with support for viewing diffs and command parameters), custom rejection reasons, underlying sandbox upgrade approval workflows, bilingual support in Chinese and English, and configuration persistence. Users can install it via the `dsh plugin` command and enable "custom review" in the permission selector. This plugin is suitable for scenarios that require clear boundaries for AI operations, and it is recommended to start from small
Read MoreFlask User Authentication: Implementing Permission Control with Flask-Login
This article introduces how to implement user authentication and permission control for web applications using Flask-Login. The core steps include: first, installing necessary libraries such as Flask, Flask-Login, Flask-SQLAlchemy, and Werkzeug. Configure the application and user model, define a User class inheriting from UserMixin, storing username, password hash, and role fields (with password encrypted using Werkzeug). Set up the user loading function to load users from the database via @login_manager.user_loader. Implement login and logout functions: verify username and password during login, then use login_user to maintain the session; use logout_user for logout. Protect routes with the @login_required decorator, and further control permissions through the role field. Key considerations: passwords must be stored encrypted, SECRET_KEY should be securely configured, and ensure the user loading function works correctly. The implementation ultimately achieves user session maintenance, route permission control, and basic role validation, with extensibility for features like "Remember Me" and OAuth.
Read More