CyberStrike v1.1.14

The biggest release since HackBrowser shipped. v1.1.14 brings a methodology engine, smarter orchestration, massive provider expansion, and a wave of stability fixes that make autonomous pentesting more reliable than ever.

Methodology Engine

CyberStrike now ships with a built-in methodology engine. Instead of blindly scanning, agents follow structured attack methodologies — prioritizing high-impact vectors, tracking coverage, and avoiding redundant work.

• /methodology dialog in TUI to inspect current methodology state
• Sidebar digest showing real-time methodology progress
• Intelligent orchestrator delegation based on methodology context

Smarter Agent Orchestration

The orchestrator got a major upgrade. It now delegates tasks to specialized sub-agents with full context awareness — understanding what's been tested, what failed, and what to prioritize next.

• 20 agent bug fixes in a single pass — context handling, prompt formatting, tool routing
• Cumulative token usage tracking across main agent + all sub-agents displayed in TUI
• Sub-agent prompt payloads capped to prevent context overflow

AI Provider & Model Support — 144 Providers, 800+ Models

This release brings official support for Anthropic and DeepSeek's latest models, along with a full upstream provider/model sync that dramatically expands what you can run CyberStrike with.

New Official Support

• Anthropic Claude — Pro/Max subscription support via OAuth (use your included API quota directly)
• DeepSeek V4 — Pro and Flash variants with thinking mode and reasoning effort control
• OpenAI GPT-5 — reasoning summary support for non-chat models
• GitHub Copilot — Enterprise support with responses API routing
• Local LLMs — Ollama, LM Studio with CLI management and UI improvements

Upstream Provider Sync

Full sync with the latest upstream provider registry — CyberStrike now supports 144 providers including:

• Major: Anthropic, OpenAI, Google (Gemini/Vertex), Azure, Amazon Bedrock, Mistral, Cohere, xAI (Grok)
• Performance: Groq, Cerebras, Fireworks AI, Together AI, DeepInfra
• Routing: OpenRouter, Requesty, Helicone, Vercel AI
• Chinese providers: DeepSeek, Alibaba (Qwen), Tencent, Zhipu AI, Kimi, Minimax, Baichuan, SiliconFlow, StepFun
• Cloud: Cloudflare Workers AI, DigitalOcean, Scaleway, OVHCloud, Vultr, Databricks, Snowflake Cortex
• New additions: Venice, Nova, Poolside, Trinity, GitLab AI Gateway, SAP AI Core, and 50+ more

CIS Benchmark Skills — 700+ New Controls

Massive expansion of compliance automation:

• Kubernetes v1.11.1, v1.12.0, v2.0.0
• Docker v1.7.0, v1.8.0
• Apache HTTP Server 2.2 (v3.6.0) and 2.4 (v2.3.0)
• Apache Cassandra 3.11, 4.0, 4.1, 5.0
• Apache Tomcat 10.1
• Ubuntu 18.04, 20.04
• Azure Foundations v5.0.0, Compute v2.0.0, Database v2.0.0, Storage v1.0.0
• Google Cloud Platform Foundation v4.0.0, Container-Optimized OS v1.2.0, Workspace v1.3.0
• MITRE ATT&CK upgraded with Atomic Red Team mappings

All benchmarks reorganized into Cloud_Providers/ and Server_Software/ taxonomy.

HackBrowser Improvements

• Auth failures now surface immediately instead of producing silent empty crawls
• Subscription parity between worker and main process
• Multi-credential headless mode fixes
• Playwright version pinned exactly in postinstall (no more caret-range drift)

Performance & Stability

• Agent init: cold-start eliminated with lightweight directory scan
• First message: 20s → <1s via lazy skill loading
• Log growth: 11GB/3min → <100MB (permission evaluate moved to DEBUG)
• Test suite: 64 pre-existing failures resolved
• Session context: tool-error strings truncated before storage
• Permission system: DeniedError messages capped instead of serializing full ruleset

False Positive Reduction

• Vulnerability deduplication by endpoint + attack vector
• Prompt hardening to block 403/public-endpoint false positives
• Session context integration for smarter assessment

Install / Upgrade

bunx @cyberstrike-io/cyberstrike@latest

Contributors: @badchars, @Serhatcck