{"slug": "cyberstrike-open-source-ai-harness-for-offensive-security-agpl", "title": "CyberStrike – open-source AI harness for offensive security (AGPL)", "summary": "CyberStrike, an open-source AI harness for offensive security released under AGPL, turns any LLM subscription into an autonomous red team agent with 13+ specialized agents, 7,600+ security skills, and 120+ OWASP test cases. The tool supports 150+ AI providers and 5,300+ models, and includes 56+ built-in tools and 176+ MCP tools. It is available via npm and offers a terminal TUI, with documentation at docs.cyberstrike.io.", "body_md": "[English](/CyberStrikeus/CyberStrike/blob/main/README.md) |\n[简体中文](/CyberStrikeus/CyberStrike/blob/main/README.zh.md) |\n[繁體中文](/CyberStrikeus/CyberStrike/blob/main/README.zht.md) |\n[한국어](/CyberStrikeus/CyberStrike/blob/main/README.ko.md) |\n[Deutsch](/CyberStrikeus/CyberStrike/blob/main/README.de.md) |\n[Español](/CyberStrikeus/CyberStrike/blob/main/README.es.md) |\n[Français](/CyberStrikeus/CyberStrike/blob/main/README.fr.md) |\n[Italiano](/CyberStrikeus/CyberStrike/blob/main/README.it.md) |\n[Dansk](/CyberStrikeus/CyberStrike/blob/main/README.da.md) |\n[日本語](/CyberStrikeus/CyberStrike/blob/main/README.ja.md) |\n[Polski](/CyberStrikeus/CyberStrike/blob/main/README.pl.md) |\n[Русский](/CyberStrikeus/CyberStrike/blob/main/README.ru.md) |\n[Bosanski](/CyberStrikeus/CyberStrike/blob/main/README.bs.md) |\n[العربية](/CyberStrikeus/CyberStrike/blob/main/README.ar.md) |\n[Norsk](/CyberStrikeus/CyberStrike/blob/main/README.no.md) |\n[Português (Brasil)](/CyberStrikeus/CyberStrike/blob/main/README.br.md) |\n[ไทย](/CyberStrikeus/CyberStrike/blob/main/README.th.md) |\n[Türkçe](/CyberStrikeus/CyberStrike/blob/main/README.tr.md) |\n[Українська](/CyberStrikeus/CyberStrike/blob/main/README.uk.md) |\n[বাংলা](/CyberStrikeus/CyberStrike/blob/main/README.bn.md) |\n[Ελληνικά](/CyberStrikeus/CyberStrike/blob/main/README.el.md) |\n[Tiếng Việt](/CyberStrikeus/CyberStrike/blob/main/README.vi.md) |\n[हिन्दी](/CyberStrikeus/CyberStrike/blob/main/README.hi.md)\n\nAutomated penetration testing from your terminal — plug in your Claude, GPT, or any LLM subscription\n\nand turn it into an autonomous red team agent with 13+ specialized agents, 7,600+ security skills, and 120+ OWASP test cases.\n\n**150+ AI providers** • **5,300+ models** • **56+ built-in tools** • **176+ MCP tools**\n\n[Quick Start](#quick-start) •\n[Intelligence Layer](#intelligence-layer) •\n[What Makes It Different](#what-makes-it-different) •\n[Agents](#agents) •\n[Skills](#security-skills) •\n[Web UI](#web-ui--remote-access) •\n[Bolt](#bolt--remote-tool-execution) •\n[MCP Ecosystem](#mcp-ecosystem) •\n[Post-Exploitation](#post-exploitation) •\n[Installation](#installation) •\n[Docs](https://docs.cyberstrike.io) •\n[Website](https://cyberstrike.io)\n\n```\nnpm i -g @cyberstrike-io/cyberstrike@latest && cyberstrike\n```\n\nThat's it. CyberStrike launches a TUI in your terminal, asks for your LLM provider and API key on first run, and you're ready to go. Tell it what to test — it handles reconnaissance, vulnerability discovery, exploitation, and reporting autonomously.\n\nAlready have a Claude Code or OpenAI subscription?CyberStrike's intelligence layer sits on top of your existing AI subscription. No separate API costs — your current plan powers an entire pentest toolkit.\n\nExplore the full documentation at ** docs.cyberstrike.io** or visit\n\n**for demos and guides.**\n\n[cyberstrike.io](https://cyberstrike.io)CyberStrike isn't just a wrapper around an LLM. It's an intelligence layer that transforms any AI model into an offensive security specialist.\n\n**How it works:** When you connect your LLM provider, CyberStrike injects domain-specific context — OWASP testing methodology, vulnerability patterns, attack chain reasoning, and tool orchestration logic — into every interaction. The model doesn't need to know security; CyberStrike teaches it.\n\n**What the intelligence layer provides:**\n\n**Schema normalization**— Structured output from any provider, regardless of response format differences** Context guard**— Prevents prompt leakage and keeps the agent focused on the current test phase** Provider auto-detection**— Automatically identifies your LLM endpoint and configures the optimal transport** Tool orchestration**— Chains security tools intelligently based on findings, not fixed scripts\n\n**150+ AI providers and 5,300+ models supported out of the box:**\n\nCyberStrike integrates with the entire AI ecosystem through 23 bundled SDK providers and 150+ providers via the [models.dev](https://models.dev) catalog. Here are the core integrations:\n\n| Provider | Models | Notes |\n|---|---|---|\nAnthropic |\nClaude 4.5, Claude 4 | Best performance with extended thinking |\nOpenAI |\nGPT-5, GPT-4.1, o3, o4 | Full tool-use + reasoning support |\nGoogle |\nGemini 2.5 Pro/Flash | Long context for large codebases |\nAmazon Bedrock |\nAll Bedrock models | IAM auth, no API keys needed |\nAzure OpenAI |\nAll Azure-hosted models | Enterprise deployments |\nGoogle Vertex AI |\nGemini + Claude on GCP | Regional endpoints (EU/US) |\nGitHub Copilot |\nGPT-5, Claude, Gemini | Use your existing Copilot subscription |\nxAI |\nGrok 3, Grok 3 Mini | Real-time data access |\nGroq |\nLLaMA, Mixtral | Ultra-fast inference |\nMistral |\nMistral Large, Codestral | European data residency |\nDeepSeek |\nDeepSeek V3, R1 | Cost-effective alternative |\nCerebras |\nLLaMA on Cerebras | Fastest inference available |\nCohere |\nCommand R+ | RAG-optimized models |\nOpenRouter |\n300+ models | Single API, any model |\nTogether AI |\nOpen-source models | Fine-tuning support |\nDeepInfra |\nOpen-source models | Pay-per-token, no GPU needed |\nPerplexity |\nSonar models | Search-augmented generation |\nAlibaba Cloud |\nQwen, Kimi, DashScope | Chinese model ecosystem |\nCloudflare AI Gateway |\nAny provider via gateway | Caching, rate limiting, analytics |\nOllama |\nAny GGUF model | Fully offline, local-only |\nLM Studio |\nAny local model | Desktop GUI + API server |\nvLLM |\nAny HuggingFace model | Self-hosted, GPU-optimized |\nAny OpenAI-compatible |\n— | Custom endpoints welcome |\n\nAir-gapped environments?Run CyberStrike entirely offline with Ollama or LM Studio. No data leaves your machine — ever.\n\n|\nCyberStrike ships with 13+ agents purpose-built for security domains. Each agent carries domain-specific methodology, tool knowledge, and testing patterns. The web-application agent follows OWASP WSTG. The cloud-security agent knows CIS benchmarks. The mobile agent uses Frida and follows MASTG/MASVS. They don't guess — they follow proven offensive security frameworks. |\nMost AI security tools are thin wrappers that send your prompt to an API. CyberStrike's intelligence layer normalizes outputs across 150+ providers and 5,300+ models, guards context between test phases, auto-detects your provider configuration, and orchestrates multi-step attack chains. The result: consistent, methodology-driven pentesting regardless of which model you use. |\n|\nAnthropic, OpenAI, Google, Amazon Bedrock, Azure, Groq, Mistral, xAI, DeepSeek, Cerebras, Cohere, OpenRouter, Together AI, GitHub Copilot — or run fully offline with Ollama and LM Studio. 150+ providers, 5,300+ models. You choose the model. You own the results. As AI models get better and cheaper, CyberStrike gets better with them. Switch providers in seconds without reconfiguring anything. |\nYour security tools don't have to run on your laptop. Deploy Bolt on one or many remote servers, pair with Ed25519 keys, and control everything from your local terminal. One CyberStrike instance can orchestrate dozens of Bolt servers — each with its own toolkit, network position, and attack surface access. |\n\nSwitch between agents with `Tab`\n\n. Each one is a domain specialist.\n\n| Agent | Focus | What It Does |\n|---|---|---|\ncyberstrike |\nGeneral | Full-access primary agent — reconnaissance, exploitation, reporting |\nweb-application |\nWeb | OWASP Top 10, WSTG methodology, API security, session testing |\nmobile-application |\nMobile | Android/iOS, Frida/Objection, MASTG/MASVS compliance |\ncloud-security |\nCloud | AWS, Azure, GCP — IAM misconfigs, CIS benchmarks, exposed resources |\ninternal-network |\nNetwork | Active Directory, Kerberos attacks, lateral movement, pivoting |\n\nPlus **8 specialized proxy testers** that run automatically on intercepted traffic:\n\n| Tester | What It Tests |\n|---|---|\nIDOR |\nObject-level access control — can user A reach user B's resources? |\nAuthorization Bypass |\nVertical privilege escalation — can low-privilege users hit admin endpoints? |\nMass Assignment |\nUnexpected writable fields — role, price, balance, userId in request bodies |\nInjection |\nSQL, command, LDAP, template injection across all input vectors |\nAuthentication |\nToken validation, session fixation, credential exposure |\nBusiness Logic |\nPrice manipulation, coupon reuse, race conditions, workflow bypass |\nSSRF |\nInternal host access via user-controlled URLs or redirect parameters |\nFile Attacks |\nPath traversal, unrestricted upload, dangerous file types |\n\nEach tester uses a **3-gate confirmation protocol**: execute a baseline request, execute the attack, compare responses. A finding is only reported when there is a measurable, reproducible difference — not on speculation. Duplicate findings (same endpoint + attack vector) are automatically suppressed across the session.\n\nCyberStrike ships with **7,600+ security skill files** — structured, Ed25519-signed methodology documents that give agents deep domain knowledge at runtime. Skills are lazy-loaded (one at a time, on demand) and statically injected into agent prompts.\n\n**Skill categories:**\n\n| Category | Skills | What They Cover |\n|---|---|---|\nAttack Methodologies |\n19 | JWT attacks, SSRF, SSTI, race conditions, request smuggling, cache poisoning, CORS, GraphQL, prototype pollution, XXE, WebSocket, subdomain takeover, host header injection, open redirect |\nPost-Exploitation |\n5 | AWS, Azure, Kubernetes, Windows, macOS privilege escalation and persistence |\nCompliance Frameworks |\n3 | CIS Benchmarks (AWS/Azure/GCP/K8s), NIST Framework, MITRE ATT&CK (Enterprise, Mobile, ICS) |\nDomain Knowledge |\n8+ | Active Directory security, web security patterns, recon methodology, CI/CD attacks, Kerberos attacks, eBPF techniques |\n\nEach skill includes testing procedures, payloads, tool commands, and CWE mappings. Skills are tagged with OWASP WSTG IDs, CIS control IDs, and chain relationships — so agents know which skills to combine for multi-step attack chains.\n\nFull documentation:\n\n[docs.cyberstrike.io/docs/tools/hacker-browser]\n\nHackBrowser is CyberStrike's built-in Chromium browser. Start it from the TUI with `/hackbrowser`\n\n. As you browse, every HTTP request is captured and routed through the proxy-agent pipeline — no manual export, no Burp project files.\n\n**Two capture modes:**\n\n**Manual**— Browse the target yourself. Log in as different users, navigate features, trigger actions. HackBrowser captures the real API traffic behind every click.**Autonomous**— Provide credentials for multiple accounts, set a scope, and let HackBrowser crawl automatically. It logs in as each user, maps reachable pages, and captures the traffic difference between roles.\n\n**Role & credential discovery:**\n\nAs you browse with multiple accounts, CyberStrike builds a session context — a live map of discovered credentials, inferred role hierarchy, and which endpoints each role can reach. The 8 proxy sub-testers use this context directly: they know which token to use for a high-privilege baseline and which lower-privilege credentials to test with, without any manual setup.\n\n```\nBrowser traffic → Proxy intercept → Orchestrator → 8 sub-testers (parallel)\n                                          ↓\n                               Session context (credentials, roles,\n                               endpoints, functions) shared across all testers\n```\n\n**Scope control:**\n\nUse `--scope`\n\nto limit testing to specific domains. CyberStrike automatically derives the registered domain (e.g. `--scope api.example.com`\n\ncovers `api.example.com`\n\nbut not `other.com`\n\n). Pass multiple `--scope`\n\nflags for multi-domain targets.\n\nCyberStrike includes a full web interface. Run `cyberstrike web`\n\nand control your agents, MCP servers, Bolt connections, and vulnerability findings from any browser.\n\n**Access from anywhere with Cloudflare Tunnel:**\n\n```\nBrowser ──HTTPS──▶ Cloudflare Tunnel ──encrypted──▶ cloudflared (localhost) ──▶ CyberStrike Server\nexport CYBERSTRIKE_SERVER_PASSWORD=your-secure-password\ncyberstrike web\n# In another terminal:\ncloudflared tunnel --url http://localhost:4096 run your-tunnel\n```\n\n**Why this is secure:**\n\n**Zero open ports**— CyberStrike binds to`localhost:4096`\n\n.`cloudflared`\n\nmakes an outbound-only connection to Cloudflare's edge. No firewall rules, no port forwarding needed.**End-to-end encryption**— Browser to Cloudflare edge is TLS. Cloudflare edge to your machine is an encrypted tunnel. No plaintext leaves your network.**Password-protected API**— Every API request requires Basic Auth. Local requests on`localhost`\n\nbypass auth for convenience; remote requests via CF tunnel always require credentials (detects`X-Forwarded-For`\n\n/`CF-Connecting-IP`\n\n).**Your data stays local**— LLM inference runs on your hardware. CyberStrike processes everything locally. The tunnel is just a secure pipe.\n\n**What's in the Web UI:**\n\n| Tab | What It Does |\n|---|---|\nChat |\nFull conversation with all 13+ security agents |\nMCP |\nLive MCP server status, health, and tool counts |\nBolt |\nBolt remote server connection monitoring |\nVulnerabilities |\nDiscovered vulns with severity, PoC, and impact |\nWeb Context |\nEndpoints, roles, credentials, and functions discovered during active sessions |\n\n** app.cyberstrike.io** is a hosted static page (no backend, no data storage) for convenience. Or self-host: clone the repo and serve\n\n`packages/app/dist/`\n\nfrom your own domain.Bolt is CyberStrike's remote tool server. Deploy it on any VPS, cloud instance, or Docker container — then control it from your local terminal over MCP protocol with Ed25519 authentication.\n\n**One CyberStrike, many Bolt servers:**\n\n```\n                                          ┌─────────────────────┐\n                                     ┌───►│  Bolt Server #1     │\n                                     │    │  nmap, nuclei, ffuf  │\n┌──────────────────┐   MCP + Ed25519 │    └─────────────────────┘\n│  Your Terminal   │   over HTTPS    │    ┌─────────────────────┐\n│  CyberStrike TUI │ ◄─────────────►├───►│  Bolt Server #2     │\n│                  │   Tool Results   │    │  sqlmap, burp, zap   │\n└──────────────────┘                 │    └─────────────────────┘\n                                     │    ┌─────────────────────┐\n                                     └───►│  Bolt Server #3     │\n                                          │  Custom toolkit      │\n                                          └─────────────────────┘\n```\n\n**Deploy anywhere**— VPS, Docker, Kubernetes, or bare metal with pre-built Kali images** Ed25519 key pairing**— No passwords, no shared secrets, no attack surface** Real-time streaming**— Results flow back to your TUI as they happen** Manage from TUI**— Add, remove, and monitor Bolt servers without leaving CyberStrike** Scale horizontally**— Run heavy scans from servers with better bandwidth while you work locally\n\nCyberStrike connects to specialized MCP servers that extend its capabilities — **176+ security tools** across 5 domains:\n\n| Server | Tools | What It Adds |\n|---|---|---|\n|\n\n[github-security-mcp](https://github.com/badchars/github-security-mcp)[cve-mcp](https://github.com/badchars/cve-mcp)[osint-mcp](https://github.com/badchars/osint-mcp)All open source. All installable with `npx`\n\n. Plug them into CyberStrike or use them standalone with any MCP-compatible client.\n\nCyberStrike agents have direct access to **56+ tools** without any external dependencies:\n\n| Category | Tools |\n|---|---|\nExecution |\nShell (bash), file read/write/edit/patch, directory listing, batch operations |\nDiscovery |\nWeb fetch, web search, code search, glob, grep, intel gathering |\nOffensive |\nHackBrowser, attack script execution, vulnerability reporting & triage |\nPost-Exploitation |\nAWS hook, Azure hook, Kubernetes hook, Windows hook, macOS hook, CI/CD pipe, eBPF |\nWeb Context |\nSession context, endpoint/role/credential/function discovery and management |\nProxy |\nHTTP/HTTPS interception, request replay, session context sharing across sub-testers |\nReporting |\nProfessional report generation, coverage notes, methodology tracking, VRT checks |\nIntegration |\nMCP servers, Bolt remote tools, custom plugins, LSP |\n\nPlus a **plugin SDK** with 15+ hook types (tool interception, message transformation, permission prompts, shell environment) — build your own agents and tools, register them at runtime.\n\nCyberStrike includes built-in post-exploitation capabilities across multiple platforms — no external tools required.\n\n| Platform | Capabilities |\n|---|---|\nmacOS |\nChrome credential extraction, Keychain dumping, keylogging, TCC bypass, GateKeeper bypass, XProtect checks, SSH key extraction, DTrace system tracing |\nWindows |\nPost-exploitation hooks for privilege escalation and persistence |\nLinux/eBPF |\n29 kernel-level scripts — process execution monitoring, SSL/TLS sniffing, keystroke logging, namespace manipulation detection, rootkit detection, process/file/connection hiding |\nAWS |\nIAM enumeration, S3 exposure, Lambda backdoors, CloudTrail evasion |\nAzure |\nIdentity enumeration, storage exposure, function exploitation |\nKubernetes |\nPod escape, service account abuse, secret extraction, RBAC exploitation |\nCI/CD |\nPipeline injection, secret extraction, build artifact manipulation |\n\nAll post-exploitation tools are agent-driven — they execute based on context and findings, not as fixed scripts.\n\n```\n# npm (recommended)\nnpm i -g @cyberstrike-io/cyberstrike@latest\n\n# bun / pnpm / yarn\nbun add -g @cyberstrike-io/cyberstrike@latest\n\n# macOS (Homebrew)\nbrew install CyberStrikeus/tap/cyberstrike\n\n# Windows (Scoop)\nscoop install cyberstrike\n\n# Linux / macOS (curl)\ncurl -fsSL https://cyberstrike.io/install.sh | bash\n```\n\n**Pentesters**— Automate the repetitive parts. Let agents handle recon and initial testing while you focus on the creative attack chains that need human intuition.**Bug Bounty Hunters**— Faster reconnaissance, wider coverage, consistent methodology across programs. CyberStrike doesn't get tired at 3am.** Security Teams**— Run structured OWASP assessments with reproducible methodology. Get reports that map to standards your compliance team understands.**Security Researchers**— Extend CyberStrike with custom agents and MCP servers. The plugin system and MCP protocol make it a platform, not just a tool.\n\nCyberStrike is built by the security community, for the security community. We welcome contributions across:\n\n**Security agents and skills**— New attack methodologies, testing patterns, vulnerability detection** MCP servers**— Connect new security tools and data sources** Knowledge base**— WSTG, MASTG, PTES, CIS methodology guides** Core improvements**— Performance, UX, provider integrations, bug fixes\n\nRead the [Contributing Guide](/CyberStrikeus/CyberStrike/blob/main/CONTRIBUTING.md) before submitting a PR. All contributions must follow the project's [ethical use policy](/CyberStrikeus/CyberStrike/blob/main/CODE_OF_CONDUCT.md) — CyberStrike is for authorized security testing only.\n\n[AGPL-3.0-only](/CyberStrikeus/CyberStrike/blob/main/LICENSE) — Free for personal and open-source use. Commercial licensing available via [contact@cyberstrike.io](mailto:contact@cyberstrike.io).\n\nCyberStrike is the core platform. These MCP servers extend its capabilities:\n\n| Project | Domain | Tools |\n|---|---|---|\nCyberStrike |\nAutonomous offensive security agent |\n13+ agents, 56+ tools, 7,600+ skills, 150+ AI providers |\n|\n\n[github-security-mcp](https://github.com/badchars/github-security-mcp)[cve-mcp](https://github.com/badchars/cve-mcp)[osint-mcp](https://github.com/badchars/osint-mcp-server)\n[ cyberstrike.io](https://cyberstrike.io) ·\n\n[·](https://docs.cyberstrike.io)\n\n**Docs**[·](https://discord.gg/snunAaHf6U)\n\n**Discord**[·](https://x.com/cyberstrikeio)\n\n**X.com**\n\n**GitHub**\nBuilt by hackers who got tired of copy-pasting between terminals.", "url": "https://wpnews.pro/news/cyberstrike-open-source-ai-harness-for-offensive-security-agpl", "canonical_source": "https://github.com/CyberStrikeus/CyberStrike", "published_at": "2026-08-22 09:44:44+00:00", "updated_at": "2026-08-22 10:14:06.450771+00:00", "lang": "en", "topics": ["ai-tools", "ai-agents", "ai-infrastructure", "ai-products"], "entities": ["CyberStrike", "Claude", "GPT", "OWASP", "models.dev"], "alternates": {"html": "https://wpnews.pro/news/cyberstrike-open-source-ai-harness-for-offensive-security-agpl", "markdown": "https://wpnews.pro/news/cyberstrike-open-source-ai-harness-for-offensive-security-agpl.md", "text": "https://wpnews.pro/news/cyberstrike-open-source-ai-harness-for-offensive-security-agpl.txt", "jsonld": "https://wpnews.pro/news/cyberstrike-open-source-ai-harness-for-offensive-security-agpl.jsonld"}}