dsh-plugin-coding-delegate: 代码执行边界与双重签名审计
dsh-plugin-coding-delegate 是 DeepSeek Harness (DSH) 的一个插件，旨在通过插件化架构为自主编程智能体提供安全控制。该插件强制分离“架构师”（设计任务）与“执行者”（本地沙箱运行代码）角色，利用 Ed25519 双签机制和 Git Notes 记录，防止规划者直接编写代码，确保代码变更的可审计性与防自批准。安装需从仓库运行脚本，典型用法包含准备合同与人工批准两阶段流程。
Read MoreAudit Context Building: Build the Audit Context Before Vulnerability Hunting
Trail of Bits' audit-context-building Skill begins the audit and threat modeling process by documenting assumptions, guarantees, and dependencies on a per-function basis. It generates a DOSSIER and function-level records, marking “nothing found” results and open questions, and does not name vulnerabilities. It is compatible with Claude Code, Codex, and similar tools, and can be installed via the trailofbits/skills marketplace.
Read Moredsh-proof: Evidence-Driven Completion Proof and Regression Attribution Plugin
`dsh-proof` is an open-source plugin designed to convert self-reports from agents in DeepSeek Harness (DSH) into verifiable evidence chains. Through content addressing, signed checkpoints, and offline anchors, it builds a tamper-resistant chain of trust, providing evidence-driven proofs of completion, regression attribution, and real-time verification. The plugin adopts a layered architecture with a core layer decoupled from the host, and includes a DSH plugin, an MCP server, and various adapters. It is applicable to safety-critical development, stringent regression testing, and third-party auditing, shifting agent development from "blind execution" to "verifiable execution."
Read Moredsh-git-push: DSH Git Automatic Commit, Push, and Audit Plugin
`dsh-git-push` is a DeepSeek Harness plugin maintained by EIGHTfs, designed to automate Git operations in AI agent development. It integrates scanning, one-click commit and push, code auditing (including 22 rule packs), and npm publishing capabilities. The plugin supports credential management, automatic selection of the remote and commit identity, and offers an HTTP API, a standalone CLI, and auto-push functionality for AI tasks, significantly reducing environment configuration costs.
Read MoreDeepSeek Harness Security Guard Plugin
`dsh-security-guard` is a security plugin for DeepSeek Harness (DSH) that uses static scanning to assess the security of third-party plugins and prevent risks such as token leakage, malicious network connections, and code obfuscation. It includes 28 built-in detection rules covering multiple scenarios, including token handling, data exfiltration, dynamic execution, obfuscation, network listening, supply chain integrity, and persistence. The plugin supports startup self-checks, DSH tool registration, alerting mechanisms (with Webhook support), a local web dashboard, and remote rule updates. To install it, exit DSH Host and use the official loading method. While the plugin is running, we recommend checking the source code and license.
Read Moredsh-license-obligation-proof: Evidence Generation Plugin for License Obligation Delivery Closed Loop
The `dsh-license-obligation-proof` plugin for DeepSeek Harness (DSH) is designed to generate and validate delivery evidence for compliant artifacts, filling the gap between scanning and proof of delivery. It focuses on offline, deterministic evidence generation and consistency validation, producing reports that include hashes and metadata, while explicitly indicating that exhaustive component-set coverage and legal compliance cannot be proven. The plugin does not perform scanning or legal interpretation and must be used in conjunction with the DSH environment.
Read MoreDeepSeek Harness Plugin: dsh-audit-mode Standalone Code Audit Mode
The `dsh-audit-mode` plugin for DeepSeek Harness (DSH) provides an independent code auditing mechanism for agent development. As the fifth preset mode, it employs a Codex, Claude Code, or DSH LLM reviewer, combined with manual review and persistent storage, to ensure secure task execution. Users can control the audit workflow through a Web or TUI interface, with support approving fixes, pausing sessions, and viewing history. Once installed, it can be enabled via a command or interface, making it suitable for development scenarios that require security auditing in production environments.
Read Moredsh-security-assurance: DeepSeek Harness Policy-Driven Repository Security Assessment Plugin
`dsh-security-assurance` is a policy-driven repository security assessment plugin for DeepSeek Harness. It integrates with the Harness API and encapsulates the assessment process into queryable, auditable, versioned results, aiming to address the difficulty of tracing the provenance of static scanning results. The plugin primarily provides Node.js project lifecycle checks, npm publishing surface checks, pnpm lockfile integrity validation, GitHub Actions permission checks, and normalization and independent verification of reports such as Gitleaks. It supports three assessment modes—REPOSITORY, CHANGE, and TARGETED—and relies on a Node.js environment. As a security assurance tool, the plugin focuses on evidence collection and independent verification rather than general-purpose scanning. It only supports
Read Moredsh-plugin-verify: Evidence Verification Toolkit for DeepSeek Harness Agents
`dsh-plugin-verify` is a read-only workflow plugin for DeepSeek Harness (DSH), designed to help agents verify their behavior and environment state through automated checks. It provides multiple verification modes: keyword-based evidence checks, configuration file validation (with JSON/YAML structure testing), HTTP resource probing, NPM package and GitHub repository checks, and batch concurrent checks. The plugin relies on heuristic search, does not support write operations, and is primarily used in code auditing, configuration validation, and resource probing scenarios.
Read MoreDeepSeek Harness Audit Mode Plugin Overview (dsh-audit-mode)
`dsh-audit-mode` is the fifth mode in DeepSeek Harness, introducing an independent, persistent Codex audit mechanism and a manual approval workflow as a safety net for the system. It supports Codex, Claude Code, or DSH itself as the reviewer, and provides dual control interfaces for both TUI and Web. During auditing, warnings do not pause the session, but critical warnings or consecutive reviewer failures trigger a pause. A final audit is executed at the end of the session, performing a full check. This plugin is designed to reduce code generation risks and supports DSH version 0.1.3-alpha.2.
Read Moreoukeming64-tech/codex-skills: Evidence-First Agent Skills for Handoff Auditing and Documentation
`codex-skills` is a community-developed, instruction-based skill set for DeepSeek Harness, designed to address edge cases in which code, evidence, and documentation become inconsistent during agent handoffs. It includes two core skills: `handoff-auditor`, which audits stated claims against the actual state before a handoff; and `docs-sync-guardian`, which enforces synchronization between documentation and code changes. The plugin is lightweight, delivered solely via the file system, and depends on neither network access nor background services, making it suitable for validating others’ handoffs or preventing documentation drift.
Read Moredsh-windows-workspace-guard: Windows Workspace and Credential Protection Plugin
`dsh-windows-workspace-guard` is a Windows agent-protection plugin developed specifically for DeepSeek Harness (DSH), designed to control dangerous file operations by agents. Independent of official components, it intercepts tool calls and provides three decision mechanisms—`PASS`, `ASK` (approval), and `HARD BLOCK` (interception)—to prevent file deletion, credential reading, or system-state corruption. It supports multiple tool calls and includes pre-check capabilities, audit logs, and summary statistics. Installation is straightforward, and users can configure `workspaceRoots` and `sensitivePaths` to isolate environments. The plugin does not send network requests; it is a conservative boundary-control tool suitable for use in Windows environments.
Read Moredsh-vulnsec-bridge: A security plugin for integrating VulnClaw and DeepSec into DeepSeek Harness
# Introduction When performing security-related tasks in DeepSeek Harness (dsh), the common practice is to open a separate terminal to run penetration testing or code auditing CLIs, then copy the output back to the session for further analysis. As tool capabilities are not registered as dsh tools, the model cannot invoke them directly, and their results cannot be integrated into subsequent automated workflows. dsh-vulnsec-bridge is a bridging plugin designed for this scenario: it wraps VulnClaw (AI penetration testing CLI) and DeepSec (Shield code auditing + Spear authorized penetration) into dsh plugins, providing 18 security tools and a graphical "Security Workbench" panel. Below, we introduce its features, installation methods, and typical usage. First, let's discuss the prerequisites: the plugin's penetration...
Read MoreDSH Project Wiki: Adding an Auditable Project-Level Wiki for DeepSeek Harness
`dsh-project-wiki` is a local-first project Wiki plugin for DeepSeek Harness, developed by yu-xin-c (MIT License). It aims to address issues such as the easy loss of critical context during development, difficulty inheriting new sessions, and the high cost and risk of outdated conclusions when re-reading chat history. The plugin captures architectural knowledge, design decisions, and troubleshooting conclusions into Markdown pages within the workspace, supporting native web viewing, editing, and full-text search (based on FTS5, without vector search). Its core features include a proposal review mechanism (Agent proposals require manual approval before storage), source fingerprint association, expiration scanning after code changes, knowledge graph generation, bidirectional linking, and complete revision history. Installation requires Node.js 22.19+ or 24+, via `np`.
Read MoreSecurStack DSH Plugin: Invoking SecurStack Scanning, Policies, and Diagnostics in DeepSeek Harness
The SecurStack DSH plugin is an adapter for DeepSeek Harness, designed to enable agent workflows to directly invoke the existing `securstack` CLI without the need to re-implement security tools. The plugin primarily provides four core functions: executing repository security scans via CLI, performing policy threshold checks, running local diagnostics, and returning structured JSON results when the CLI supports it. It reuses existing SecurStack authentication methods (such as API Key or command-line login) and strictly adheres to conservative boundaries; version 1 does not include low-level logic such as scanning engines, encryption, or Shielding operations. Ensure the environment is Node.js 20 or higher before installation and enable it via the `dsh plugin` command. The plugin is suitable for...
Read MoreVariant Analysis: Locate a Bug and Eliminate All Similar Ones
Trail of Bits’ variant-analysis has turned the daily variant analysis work of security engineers into an Agent Skill: after confirming a vulnerability, it extracts the root cause, calibrates with precise patterns, searches the entire codebase while relaxing only one condition at a time, and finally triages and leaves CI rules. This article explains the five-step workflow, the selection of ripgrep, Semgrep and CodeQL, as well as the installation and invocation methods of Claude Code, Codex and Cursor, based on the official SKILL.md, plugin README and policy documents.
Read MoreDifferential Review: Trail of Bits formalizes PR security differential review into an executable workflow
Differential-review is an Agent Skill released by Trail of Bits, which conducts security-focused reviews on PRs, commits and diffs: it classifies tasks by risk rather than line count, uses git history to detect regressions, quantitatively calculates the blast radius, treats missing tests as a higher risk factor, and enforces the output of a Markdown report with line numbers and attack scenarios. This article checks the official SKILL.md, plugin README and documentation website, explaining what problems it solves, how to install it in Claude Code / Codex / the general Skill catalog, as well as three official usage scenarios and inapplicable scenarios.
Read More