dsh-llm-verifier avatar

dsh-llm-verifier

Best-of-3/5 LLM-as-a-Verifier plugin for DeepSeek Harness

PluginDeveloperAI ModelsCode reviewTerminal / Shell
Verification
L2 · Structured
Security
Medium
Health
Active
Trust
Silver

What it doesAI

Runs 3 or 5 isolated coding candidates, executes project tests, and uses an LLM verifier to select the best patch, applying it only after user authorization.

  • Supports 3 or 5 candidates with optional validation commands
  • Creates detached Git worktrees and runs candidates in parallel
  • Only applies winning patch after second explicit user authorization

AI-generated from the repo README — for reference only.

Installation

dsh plugin --profile web add github:Web0926/dsh-llm-verifier

Install method: GitHub · not yet tested in container (L3+)

Compatibility

DSH VersionStatus
not statedDeclared — not tested

Requirements

  • • Node.js: not stated
  • • DSH: declared "not stated"
  • • External credentials: none detected

Security Report

Automated static scan, not manual review.

DSH.SO AUDITMEDIUM8921b1
Automated review · daily
MEDIUM

Critical findings in non-blocking categories (dynamic code execution, shell execution, install scripts, obfuscation). Common in CLI/terminal plugins but worth reviewing.

4 critical·0 warning·5 info·11 files scanned
code-exec ×4
plugin version0.1.0dsh manifest
scanned commit8921b11732e72026-08-20
latest commit8921b11732e7

The scan result is valid for the scanned commit. New commits within 7 days are tolerated (the rating still counts); after 7 days without a rescan the badge shows outdated.

Disclaimer: automated static analysis, not a security guarantee. Always review what you install.

Second-opinion scan (vet · parallel evaluation)
vet: clean

vet verdict: vet: clean · score 100

  • R9 · info [heuristic] — 无出口常驻循环(含 await,可能是合法服务循环;交由 LLM 审计复核上下文)
  • R9 · info [heuristic] — 无出口常驻循环(含 await,可能是合法服务循环;交由 LLM 审计复核上下文)
  • R2 · info [likely] — require()(客户端加载器 factory 形参注入,DSH bundle 标准写法)
  • R2 · info [likely] — require()(客户端加载器 factory 形参注入,DSH bundle 标准写法)
  • R2 · info [likely] — require()(客户端加载器 factory 形参注入,DSH bundle 标准写法)

vet is advisory and does not change dsh’s four-tier level.

Risk: Medium4 critical · 0 warning · 5 info

• Static heuristic scan: done (11 files)

• Dependency vulnerabilities: requires deep audit (L3+)

• Permission sandboxing: requires runtime testing (L4+)

High-risk findings · 4 / 4

  • criticalChild process module usage (Node.js)src/git.ts:1
    import { execFile } from "node:child_process";
  • criticalChild process module usage (Node.js)src/process.ts:1
    import { spawn } from "node:child_process";
  • criticalChild process module usage (Node.js)src/process.ts:71
    const childProcess = spawn(request.executable, [...request.arguments], {
  • criticalChild process module usage (Node.js)src/process.ts:89
    if (terminationStarted || childProcess.pid === undefined) {

Heuristic static scan — may produce false positives. Review the source yourself before trusting.

Activity

Last commit 2026-08-20 · activity: Active

• Repo created: 2026-08-20

• Stars: ★ 3 · Forks: 0

• Health: Active — committed within last 30 days

Source

GitHub: github.com/Web0926/dsh-llm-verifier

Was this page helpful?