dsh-audit-bundle
MCP server for auditing DeepSeek Harness evidence. Enables inspection and verification of content-addressed audit manifests with fail-closed validation.
README
dsh-audit-bundle
Content-addressed audit indexes across independent DeepSeek Harness evidence producers.
This plugin is not an SBOM scanner, signer, audit logger, policy engine or archive. Existing tools already scan dependencies and individual 2Origin plugins already produce release, runtime, recovery, lineage and policy evidence. The missing layer is a small verifier that proves a particular subject/revision has enough pinned evidence from allowed, independent producers to cover declared controls.
Contract
An explicit manifest declares:
- one subject ID and revision;
- required controls with minimum eligible evidence, minimum distinct producers and allowed evidence types;
- evidence files pinned by SHA-256;
- JSON Pointers that bind every evidence file to the subject and revision;
- value-hash assertions, so expected or observed values never enter the audit index.
Verification fails closed for missing, stale or invalid JSON evidence, subject/revision mismatch, failed assertions, disallowed types, insufficient evidence or insufficient independent producers. The output contains IDs, types, producers, paths into JSON, hashes, statuses, coverage and a deterministic SHA-256 pair-tree Merkle root. It never copies evidence bodies or assertion values.
Files must be workspace-relative regular files. Symlinks, path escape, oversized input and excessive structure are rejected. The plugin performs no network calls or child processes and writes only a content-addressed JSON index under the explicit artifactDir, followed by read-back verification.
CLI
node bin/dsh-audit-bundle.mjs inspect --workspace examples/basic --manifest audit.manifest.json
node bin/dsh-audit-bundle.mjs verify --workspace examples/basic --manifest audit.manifest.json --artifactDir artifacts
The CLI emits one JSON object. A failed audit verdict exits 2; invalid usage exits 1.
DeepSeek Harness and MCP
The DSH bundle registers dsh_audit_bundle_inspect and dsh_audit_bundle_verify. The companion stdio MCP server registers audit_bundle_inspect and audit_bundle_verify through .mcp.json.
dsh plugin --profile audit-bundle add github:dongsheng123132/dsh-audit-bundle#<commit>
dsh --profile audit-bundle --dump-config
Verification
npm ci
npm test
npm run check
npm run smoke:mcp
DSH_CHECKOUT=/path/to/built/deepseek-harness npm run smoke:dsh
python C:/Users/ZhuanZ/.codex/skills/.system/plugin-creator/scripts/validate_plugin.py .
CI runs on Ubuntu and Windows. Node.js 22 or newer. MIT licensed.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。