跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

perseus-start

Use when you want to run a full, automated penetration test from start to finish (Scan -> Audit -> Exploit -> Report)

kaivyy · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add kaivyy/perseus --skill perseus-start

技能说明


name: perseus-start description: Use when you want to run a full, automated penetration test from start to finish (Scan -> Audit -> Exploit -> Report)

Perseus Start (Automated Security Assessment)

Context & Authorization

IMPORTANT: This skill orchestrates a complete security assessment on the user's own codebase. This is defensive security testing - the user is proactively finding vulnerabilities in their own code before attackers do.

Authorization: The user has explicitly invoked this assessment on their own codebase in their own development environment. This is equivalent to running a comprehensive test suite, static analyzer, or security audit tool.

This is NOT an attack on external systems. All analysis is performed on local files the user owns and controls.


Overview

This master skill orchestrates the entire Perseus security assessment lifecycle with smart auto-detection of languages, frameworks, and technologies.

Goal: Zero-touch automated security assessment with professional-grade output.

Agent Execution Strategy (Dual-Mode)

When orchestrating Perseus, adapt to the runtime capabilities:

  • Mode A — Subagents Available (Multi-Agent/Parallel): If subagent tools (e.g. Task, Agent, invoke_subagent) are available, launch phase agents and specialist deep-dives in parallel.
  • Mode B — Direct Execution (Single-Agent Fallback — UNIVERSAL): If no subagent tool exists in your runtime, DO NOT attempt to call non-existent subagent tools. Execute each phase directly in sequence: Phase -1 (Setup) ➔ Phase 0 (Detect) ➔ Phase 1 (Scan) ➔ Phase 2 (Audit) ➔ Phase 2.5 (Specialists) ➔ Phase 3 (Exploit Verify) ➔ Phase 4 (Report). Ensure mkdir -p deliverables is run first and write all files directly into deliverables/.

Engagement Mode (Required)

Before running phases, select one mode:

ModeEnvironmentBehavior
PRODUCTION_SAFELive productionPassive-first analysis and minimal safe verification only
STAGING_ACTIVEStaging/pre-productionActive safe payload testing with request throttling
LAB_FULLIsolated labFull dynamic verification and payload mutation
LAB_RED_TEAMDedicated lab environmentMulti-step adversarial simulation with kill-switches

Default mode is PRODUCTION_SAFE unless user explicitly confirms staging/lab authorization.

Smart Auto-Detection

Before starting the assessment, Perseus automatically detects:

Language Detection

FilesLanguage
package.json, *.ts, *.jsJavaScript/TypeScript
go.mod, *.goGo
composer.json, *.phpPHP
requirements.txt, *.pyPython
Cargo.toml, *.rsRust
pom.xml, *.javaJava
Gemfile, *.rbRuby
*.csproj, *.csC#

Framework Detection

Files/PatternsFramework
next.config.*, app/ directoryNext.js
nuxt.config.*Nuxt.js
angular.jsonAngular
vite.config., svelte.config.Vite/Svelte
gin import, echo importGo (Gin/Echo)
artisan, laravelPHP (Laravel)
manage.py, djangoPython (Django)
fastapi importPython (FastAPI)
actix-web, axum in Cargo.tomlRust (Actix/Axum)
spring-bootJava (Spring)
railsRuby on Rails

Infrastructure Detection

FilesTechnology
Dockerfile, docker-compose.ymlDocker
.github/workflows/*.ymlGitHub Actions
.gitlab-ci.ymlGitLab CI
*.tfTerraform
k8s/, kubernetes/, *.yaml with apiVersionKubernetes
serverless.ymlServerless
vercel.jsonVercel

API Detection

PatternsType
/graphql, schema.graphql, *.gqlGraphQL
WebSocket, ws://, wss://WebSocket
*.proto, grpcgRPC
openapi, swaggerREST/OpenAPI

AI/LLM Detection

PatternsTechnology
openai, anthropic, langchainLLM Integration
vector store, embeddingsRAG System
prompt, completionAI Features

Complete Capability Matrix

Core Phases (Always Run)

PhaseSkillPurpose
1scanMap architecture, entry points, attack surface
2auditAnalyze all vulnerability classes
3exploitVerify findings with safe PoCs
4reportGenerate executive security report

Specialist Deep-Dives (Run When Detected)

SkillTrigger ConditionExtended Coverage
apiREST/GraphQL/WebSocket/gRPC+OAuth, Cache, multi-lang
injectionNoSQL/Templates/Commands+Log4j, SSTI, multi-lang
cryptoJWT/Encryption/Hashing+multi-lang patterns
supply-chainPackage manifests+multi-lang, typosquatting
fileFile uploads/operations+Zip Slip, XXE, multi-lang
logicPayment/Auth/AI flows+AI prompt injection
clientReact/Vue/Angular/SSR+Server Components, Actions
configAlways+Docker, CI/CD, Cloud, K8s

Execution Flow

Phase -1: Engagement Setup

Action: Determine mode and boundaries

1. Detect runtime context (production/staging/lab)
2. Ask for explicit authorization scope if context is unclear
3. Set mode: PRODUCTION_SAFE, STAGING_ACTIVE, LAB_FULL, or LAB_RED_TEAM
4. Create deliverables/engagement_profile.md with:
   - mode
   - in-scope targets
   - excluded systems
   - request-rate limits
   - approved test window
   - kill-switch thresholds (error rate, latency, saturation)

Announce: "Engagement mode set to: [MODE]"


Phase 0: Auto-Detection

Action: Detect project technologies

# Initialize directories and detect technology manifests
mkdir -p deliverables outputs/schemas
ls -d package.json go.mod composer.json requirements.txt pyproject.toml Cargo.toml pom.xml build.gradle Gemfile *.csproj Dockerfile 2>/dev/null || true
1. Scan for package manifests:
 
阅读完整来源 ↗