DeepSeek Harness 插件

skill-security-guard

DeepSeek Harness community Bundle for the skill-security-guard static scanner(英文原文)

跳到安装方式

来源信息

GitHub 仓库
rrrrrredy/skill-security-guard
最近更新
2026年8月17日
分类
安全与权限
GitHub stars
1
载体类型
plugin
包路径
integrations/deepseek-harness
目录证据
上游声明已找到 dsh.bundle
证据路径
integrations/deepseek-harness/package.json#dsh.bundle
核对版本
0.1.0-rc.8
上游核对日期
2026-08-20

该证据由上游目录提供。本站没有安装、运行或安全审核这个插件。

安装

默认先复制一段 Prompt,让 Agent 读 GitHub 仓库和源码;需要自己装时再切到命令。

复制这段 Prompt,发给 DSH、Codex 或其他 Agent,让它先读 GitHub 仓库和源码。

请先不要安装或执行任何命令。阅读这个插件的 GitHub 仓库、README 和关键源码,然后用清楚、直接的方式回答以下问题,帮助我判断它是否适合我的需求:

1. 这个插件是什么,解决什么问题;
2. 适合哪些用户和典型使用场景;
3. 安装后如何使用,并给出一个最小使用示例;
4. 有哪些已知限制,以及隐私、安全、兼容性或维护风险;
5. 给出“推荐 / 有条件推荐 / 不推荐”的明确建议和理由。

请区分仓库明确说明、根据源码推断和未知信息。证据不足时请明确说明,不要猜测或照抄 README。

GitHub:https://github.com/rrrrrredy/skill-security-guard/tree/HEAD/integrations/deepseek-harness
插件名:skill-security-guard
作者:rrrrrredy

检查来源文件

安装前先看这个插件目录里的 README 和其他文件。

文件资源管理器2 个文件
README.md来源说明 · 只读预览

dsh-skill-security-guard

Community DeepSeek Harness Bundle for skill-security-guard. It registers the existing static-analysis skill and packages its Python scanner as a resolvable skill resource.

This is a community plugin, not an official DeepSeek plugin.

Compatibility

  • DeepSeek Harness / @deepseek-ai/dsh: 0.1.0-rc.6
  • Node.js: 22.19.x or 24+ (matching the package engines declaration)
  • Python: 3.10+

Install

Install the public npm package into a Harness profile:

dsh plugin --profile headless add dsh-skill-security-guard@0.1.0
dsh --profile headless --dump-config

For a local release candidate, replace the npm specifier with the path to the packed .tgz.

Use the npm package or a reviewed tarball. A github: install is intentionally unsupported because generated lib/ and assets/ are not committed, and this package does not request permission to execute a build during installation.

Use

Run a one-shot task and ask the agent to use skill-security-guard before installing or trusting an agent skill:

dsh --profile headless "Use skill-security-guard to scan ./path/to/a-skill and explain every confirmed finding."

For the browser surface, install the Bundle into the separate web profile and start it:

dsh plugin --profile web add dsh-skill-security-guard@0.1.0
dsh --profile web

The skill instructs the agent to run the packaged scripts/scan.py; it does not add a new model tool or silently scan unrelated files.

Uninstall

dsh plugin --profile headless remove dsh-skill-security-guard
dsh --profile headless --dump-config

After removal, the skill-security-guard provider and its catalog entry are absent.

Permissions and privacy

  • The Bundle itself performs no network requests and has no telemetry.
  • The scanner reads only the target supplied by the user or agent. Public URL input is fetched only when explicitly passed to the scanner.
  • The Bundle adds no upload path of its own. In an LLM-backed Harness profile, skill instructions, shell commands, and scanner output can still be sent to the model provider configured for that profile as part of normal agent operation.
  • This is static analysis, not runtime monitoring, sandboxing, provenance verification, or a substitute for human review.

Reproduce the package tests

From this directory:

pnpm install --frozen-lockfile
pnpm verify
pnpm pack --pack-destination .pack

pnpm verify builds assets from the canonical repository files, checks their SHA-256 manifest, loads and disposes the provider through real Cordis services, and runs the packaged scanner against safe and malicious fixtures.

The deterministic full-Harness test additionally needs an installed DSH entry, a packed tarball, Python, and an explicit scratch root:

DSH_ENTRY=/absolute/path/to/@deepseek-ai/dsh/lib/bin.js \
DSH_TARBALL=/absolute/path/to/dsh-skill-security-guard-0.1.0.tgz \
DSH_E2E_ROOT=/absolute/path/to/scratch \
PYTHON_EXECUTABLE=/absolute/path/to/python \
pnpm test:e2e:dsh

To verify the package from the public npm registry instead, set DSH_PACKAGE_SPEC=dsh-skill-security-guard@0.1.0 and omit DSH_TARBALL. Exactly one installation source is required.

PowerShell equivalent:

$env:DSH_ENTRY = "D:\path\to\@deepseek-ai\dsh\lib\bin.js"
$env:DSH_TARBALL = "D:\path\to\dsh-skill-security-guard-0.1.0.tgz"
$env:DSH_E2E_ROOT = "D:\path\to\scratch"
$env:PYTHON_EXECUTABLE = "C:\path\to\python.exe"
pnpm test:e2e:dsh

It installs the selected package source into a fresh headless profile, drives skill and the platform shell through a loopback-only deterministic DeepSeek protocol server, verifies the packaged scanner returns rating A, checks the append-only session JSONL for structural tool/call and tool/result evidence, and removes the Bundle with no remaining package reference in --dump-config. Successful artifacts are deleted by default; set DSH_E2E_KEEP=1 to retain the isolated profile for local inspection. The mock test does not replace the separate real-model release smoke.

Versioned candidate and public-release results are recorded in the public 0.1.0 release evidence. The release was not treated as complete until every external gate in that record passed.