Changelog

All notable changes to Cyberstrike CLI are documented here. For the full release history with downloadable assets, see GitHub Releases.

v1.1.15

• Added eBPF post-exploitation tool — 10 kernel-level programs for the internal-network agent, executed via ebpf <program> after gaining root on Linux targets
• Added Credential harvesting: pam_sniff (PAM uprobe — cleartext SSH/sudo/su passwords), ssl_sniff (SSL uprobe — TLS plaintext capture), keylog (TTY kprobe — keystroke capture)
• Added Stealth operations: proc_hide (hide processes from ps/top/htop), file_hide (hide files from ls/find), conn_hide (hide connections from netstat/ss)
• Added Monitoring: execve_sniff (system-wide process execution tracing), dns_sniff (kernel-level DNS query capture), dep_scan (runtime dependency and vulnerable library scanner)
• Added Cleanup: cleanup (enumerate and remove all CyberStrike eBPF programs from target)
• Added eBPF blind spot monitors — 20 kernel-level detection programs for attack primitives that bypass classical syscall hooks and operate through kernel subsystems invisible to standard monitoring
• Added Syscall bypass: io_uring_sniff (io_uring SQE submission monitoring — CONNECT/READ/WRITE/OPENAT via ring buffer bypass, kernel 5.1+)
• Added Fileless execution: memfd_exec (memfd_create + execveat AT_EMPTY_PATH correlation — diskless payload delivery detection)
• Added Process injection: ptrace_sniff (ATTACH → POKEDATA → SETREGS → CONT injection sequence detection), crossmem_sniff (process_vm_writev/readv cross-process memory injection)
• Added Exploit primitives: userfaultfd_sniff (userfaultfd race condition timing primitive detection)
• Added Integrity verification: bpf_integrity (bpf() syscall monitoring + bpftool baseline comparison — detect unauthorized BPF program loads, CyberStrike hook tampering)
• Added Network manipulation: netlink_sniff (netlink socket message monitoring — route/firewall rule injection detection)
• Added Sandbox evasion: seccomp_sniff (prctl/seccomp self-modification — sandbox weakening, process name masquerading, privilege restriction bypass)
• Added Memory IPC: mmap_sniff (shared memory via mmap MAP_SHARED/shmget/shmat — covert IPC without syscalls after mapping)
• Added Zero-copy transfers: zerocopy_sniff (splice/tee/sendfile64 fd-to-fd data movement invisible to buffer profilers)
• Added VDSO side-channels: vdso_sniff (clock_gettime/gettimeofday high-frequency timing + mprotect VDSO page tampering)
• Added Kernel keyring: keyring_sniff (add_key/keyctl/request_key — credential storage in kernel keyring evading filesystem monitoring)
• Added Namespace escape: namespace_sniff (setns/unshare — container escape, namespace pivoting, single-namespace monitoring bypass)
• Added Terminal injection: ioctl_sniff (TIOCSTI keystroke injection, TIOCLINUX, TIOCSCTTY terminal steal — ioctl blind spot)
• Added Mount manipulation: mount_sniff (overlay/bind mounts over /etc, /usr, /bin + FUSE mount detection)
• Added FUSE hijacking: fuse_sniff (/dev/fuse open + fuse-type mount — file operations bypass kernel VFS)
• Added Perf side-channels: perf_sniff (perf_event_open — cache miss/branch misprediction hardware counter abuse)
• Added BPF map covert channels: bpfmap_sniff (MAP_CREATE/UPDATE/LOOKUP/DELETE — inter-process data sharing via BPF maps)
• Added Dynamic linker injection: ldpreload_sniff (LD_PRELOAD env injection + ld.so.preload/conf write detection)
• Added Futex covert channels: futex_sniff (WAIT/WAKE timing-based signaling between processes, busy-wait exploitation)
• Added ebpf-attacks skill — kill chain methodology with 5 phases: situational awareness, credential harvesting, stealth operations, advanced evasion detection, cleanup. Includes MITRE ATT&CK mappings (T1014, T1040, T1055.008, T1055.012, T1056.001, T1068, T1553, T1556, T1562.001, T1562.004, T1620) and detection considerations
• Added Windows post-exploitation tool (winhook) — 12 userland programs for the internal-network agent, executed via winhook <program> after gaining Administrator on Windows targets
• Added AV/EDR evasion: amsi_bypass (patch AmsiScanBuffer in-memory), etw_blind (patch EtwEventWrite to blind EDR), defender_exclude (add Windows Defender exclusion paths)
• Added Credential harvesting: lsass_dump (LSASS memory dump via comsvcs.dll/MiniDumpWriteDump), sam_dump (SAM/SYSTEM/SECURITY registry hive extraction), dpapi_extract (DPAPI secret decryption — browser passwords, WiFi, Vault), credential_prompt (fake CredUI dialog), keylog_win (SetWindowsHookEx keystroke capture), clipboard_sniff (clipboard monitoring)
• Added Monitoring: etw_process (process creation tracking), etw_network (network connection tracking)
• Added Cleanup: cleanup_win (event log clearing, artifact removal, Defender exclusion rollback)
• Added macOS post-exploitation tool (machook) — 12 programs for the internal-network agent, executed via machook <program> after gaining root on macOS targets
• Added Credential harvesting: keychain_dump (Keychain password extraction via security CLI), chrome_creds (Chrome/Safari credential decryption — PBKDF2 + AES-128-CBC), ssh_keys (SSH private key discovery for all users), tcc_bypass (TCC.db manipulation for camera/mic/FDA access), keylog_mac (CGEventTap keystroke capture)
• Added Monitoring: dtrace_exec (process execution tracing), dtrace_net (network connection tracing), dtrace_file (file access tracing)
• Added Stealth: xprotect_check (XProtect/MRT/Gatekeeper/SIP/EDR enumeration), gatekeeper_bypass (quarantine xattr removal), log_clear (unified log, ASL, audit log clearing)
• Added Cleanup: cleanup_mac (LaunchAgent/Daemon removal, process cleanup, temp file removal)
• Added windows-postexploit skill — kill chain with AV/EDR evasion → credential harvesting → monitoring → cleanup phases. MITRE ATT&CK mappings (T1003, T1056.001, T1059.001, T1562.001, T1070.001, T1555)
• Added macos-postexploit skill — kill chain with situational awareness → credential harvesting → monitoring → stealth → cleanup phases. MITRE ATT&CK mappings (T1555.001, T1056.001, T1059.004, T1562.001, T1070.002, T1553.001)
• Added AWS post-exploitation tool (awshook) — 10 cloud programs for internal-network and cloud-security agents, executed via awshook <program> with valid AWS credentials
• Added IAM exploitation: iam_enum (IAM users/roles/groups enumeration + privilege escalation path analysis), iam_privesc (PassRole, AssumeRole, AttachPolicy, CreateAccessKey chains)
• Added Data exfiltration: s3_dump (sensitive file discovery via pattern matching + download), secrets_dump (Secrets Manager + SSM Parameter Store extraction), ec2_snapshot (EBS volume snapshot + cross-account sharing)
• Added Persistence: lambda_backdoor (layer injection or new backdoor function with admin role)
• Added Remote execution: ssm_exec (SSM RunCommand on managed EC2 instances)
• Added Credential harvesting: metadata_harvest (EC2 IMDSv1/v2, ECS task metadata, Lambda env credential extraction)
• Added Defense evasion: cloudtrail_blind (stop trails, manipulate event selectors, delete logs)
• Added Cleanup: cleanup_aws (restore CloudTrail, delete Lambda/IAM/EBS artifacts, clean state file)
• Added Azure post-exploitation tool (azurehook) — 8 cloud programs for internal-network and cloud-security agents, executed via azurehook <program> with Azure access tokens
• Added Entra ID exploitation: entra_enum (users, groups, apps, service principals, directory roles via Graph API), entra_privesc (OAuth2 consent grant, PIM role activation, SP credential injection)
• Added Data exfiltration: keyvault_dump (secrets, keys, certificates from all accessible Key Vaults), storage_dump (Blob storage sensitive file discovery + download)
• Added Credential harvesting: managed_identity (IMDS token harvest from VM/App Service for ARM, Graph, KeyVault, Storage, SQL), azuread_token (FOCI client ID abuse, token refresh, JWT decode)
• Added Persistence: runbook_backdoor (Automation Account Python3 runbook with callback + hourly schedule)
• Added Cleanup: cleanup_azure (revoke consent grants, remove SP secrets, delete runbooks/schedules)
• Added Kubernetes post-exploitation tool (kubehook) — 7 programs for internal-network and cloud-security agents, executed via kubehook <program> with valid kubeconfig
• Added Enumeration: k8s_enum (namespaces, pods, services, secrets, RBAC, nodes — 11 resource categories)
• Added Credential harvesting: k8s_secrets (Secret extraction + base64 decode across namespaces, TLS cert/dockerconfig/SA token parsing), etcd_dump (direct etcd connection for protobuf-encoded secret extraction)
• Added Privilege escalation: k8s_privesc (SA token theft, ClusterRoleBinding creation, TokenRequest API minting)
• Added Container escape: k8s_escape (privileged mode, hostPID, hostNetwork, Docker socket, cgroup release_agent — 7 detection vectors)
• Added Persistence: k8s_backdoor (privileged DaemonSet on all nodes or CronJob with callback, deployed to kube-system)
• Added Cleanup: cleanup_k8s (state file + label selector app=cyberstrike resource removal)
• Added CI/CD pipeline attack tool (cipipe) — 5 programs for the internal-network agent, executed via cipipe <program> with platform API tokens
• Added Secret extraction: gh_secrets (GitHub Actions secret enumeration, workflow log credential scanning, workflow dispatch exfiltration), gitlab_tokens (CI/CD variables, runner tokens, deploy tokens, project access tokens), jenkins_creds (credential API dump + Groovy Script Console extraction with password/SSH key/secret decryption)
• Added Pipeline injection: pipeline_inject (GitHub Actions / GitLab CI workflow file injection with env exfiltration to callback URL)
• Added Cleanup: cleanup_ci (GitHub/GitLab branch deletion from state file)
• Added aws-postexploit skill — 6-phase kill chain: recon → IAM privesc → data access → persistence → defense evasion → cleanup. MITRE ATT&CK mappings (T1078.004, T1530, T1537, T1562.008, T1098)
• Added azure-postexploit skill — 5-phase kill chain: Entra ID recon → privilege escalation → secret extraction → persistence → cleanup. MITRE ATT&CK mappings (T1078.004, T1552.001, T1098.001, T1550.001)
• Added k8s-postexploit skill — 5-phase kill chain: cluster recon → secret extraction → privilege escalation → persistence → cleanup. MITRE ATT&CK mappings (T1611, T1552.007, T1613, T1610)
• Added cicd-attacks skill — 4-phase kill chain: enumeration → secret extraction → pipeline injection → cleanup. MITRE ATT&CK mappings (T1195.002, T1552.004, T1059)
• Added GitHub Copilot Enterprise provider — verified and validated full Copilot provider support for Enterprise license holders. Use Claude, GPT, and Gemini models at zero cost through GitHub Copilot. Includes OAuth device flow auth, Enterprise Server URL support, Chat + Responses API, reasoning, tool calling, vision, and streaming. Authenticate via /provider add → github-copilot.
• Added DAST proxy-testing memory (coverage notes) — testers record what they tested per asset via record_coverage_note, scoped wide (deployment/account-wide, e.g. auth mechanism) or local (per-endpoint); the orchestrator and testers read app-wide coverage to skip redundant re-testing. New get_coverage_notes tool + coverage_note store.
• Added Per-credential observed values — REST/GraphQL/JSON-RPC requests record the concrete ID/field values seen for each credential as raw IDOR/access-control substrate, surfaced to the orchestrator and request tools and shown in the TUI/Web object tree.
• Added GraphQL & JSON-RPC operations as first-class endpoints — each operation is keyed by a deterministic value-stripped op-key, so it is modeled and tested as its own endpoint instead of collapsing onto the shared transport URL.
• Added web_get_detail tool — pull the full detail of a single object/function/role/credential on demand, complementing the names-only session-context catalog.
• Added Vulnerability triage lifecycle + grouped views — report_vulnerability now always records every finding with a status (new/approved/duplicate); triage_vulnerability links duplicates to their canonical instead of discarding them. TUI and Web group findings by status (New / Duplicate / Approved, sorted by severity) with duplicate→canonical references.
• Added Hybrid report generation — generate_report produces HackerOne-ready reports across TUI and Web surfaces, including a per-endpoint coverage subsection.
• Added Bounded session context — web_get_session_context no longer returns all accumulated session data on every call (which grew O(N²) and stalled long runs). It now returns endpoint-scoped recent objects + values inline, a names-only catalog of everything else, the last few requests, and pull tools — keeping per-call size at a few KB regardless of session size (previously ~20 KB and growing) while preserving the model’s recall of findings.
• Fixed Endpoint over-fragmentation — form-urlencoded and multipart POST bodies are value-stripped like JSON when computing the endpoint key, so the same logical endpoint hit with volatile body values (e.g. WebSocket channel-auth) collapses to one record instead of hundreds.
• Fixed Phantom hosts during normalization — //-relative targets and HTTP/2 :authority no longer produce spurious host entries.
• Fixed Silent vulnerability loss — the previous dedup path dropped roughly half of reported findings; findings are now always recorded and de-duplicated through the triage lifecycle (measured drop rate 0).
• Fixed Proxy-flow ordering — the per-endpoint analyzer runs as a blocking first step before the tester subagents, so testers act on its findings instead of racing it.
• Fixed Hackbrowser crawl coverage & robustness — broader DOM capture (shadow-DOM / web-component controls, opacity:0 design-system inputs, data-hx-* / data-ng-click framework bindings), ephemeral framework ids rejected as click selectors (crawl-loop fix), and crawl-unblock fixes (popup-trigger expand, non-blocking panel, overlay retry).
• Fixed Hackbrowser provider routing — the crawl worker routes each model through the shared provider map, so non-OpenAI keys (Gemini, Bedrock, Cohere) reach the correct API instead of silently hitting OpenAI.
• Fixed Token cost accounting — stopped double-charging reasoning tokens; aborted, errored, or context-capped subagents are no longer recorded as successful.
• Added Web UI v1.1.6-beta — branding, auth, side panel, offensive tooling (#26)
• Added Web UI bundled in npm package — auto-installs to ~/.cyberstrike/web/ via postinstall
• Added MCP tab in status popover — live MCP server status from TUI
• Added Bolt tab in status popover — live Bolt connection status from TUI
• Added MCP/Bolt config persistence — save to global scope via REST API
• Added npm-optimized README — bundled in published package for npmjs.com display
• Added Web UI build step in publish workflow
• Added Long-running task strategy added to cyberstrike agent prompt
• Added Offensive security agent prompts hardened
• Fixed CORS and auth failures on remote/tunnel access
• Fixed MCP/Bolt status fetch on panel mount + bootstrap hardening
• Fixed Enterprise infra made conditional on CYBERSTRIKE_ENTERPRISE env var
• Fixed Stripe/PlanetScale providers made optional, RegionalHostname skipped
• Fixed XDG data path for web UI install in postinstall
• Fixed PUBLISH_TOKEN PAT fallback in publish workflow
• Added Enterprise infrastructure no longer loaded by default — opt-in via env
• Added Intelligence layer, SEO optimization, and Bolt 1:N architecture in README rewrite
• Added Offensive security agent prompts and hardened publish script
• Added --beta flag for install script
• Fixed Schema reconciler to repair partially applied migrations
• Fixed @opentui/core and @opentui/solid updated to 0.1.88
• Fixed Auto-fallback to available port when default port is busy
• Fixed Scoped package names in uninstall script
• Fixed npm install commands and domain references on website
• Fixed Billing header for OAT token auth on Sonnet/Opus models
• Fixed All cyberstrike.us references replaced with cyberstrike.io
• Fixed Bin launcher script renamed opencode → cyberstrike
• Added npm scope renamed @cyberstrikeus → @cyberstrike-io
• Added License consolidated to AGPL-3.0-only, copyright year updated
• Added Docker removed from build pipeline
• Added Tauri desktop app removed (opencode legacy)
• Added 13 unused workflows inherited from opencode removed
• Added Playwright e2e removed from test workflow
• Added Bolt remote tool server — full client integration with Ed25519 key pairing, SDK fetch chain, live sidebar status
• Added Bolt management from TUI — add, delete (Ctrl+D), configure Bolt connections
• Added MCP server management — add/remove MCP servers from TUI with duplicate detection and validation
• Added Local LLM provider support — connect to any OpenAI-compatible endpoint (vLLM, Ollama, LM Studio)
• Added 23-language README — complete rewrite with social preview SVGs (dark + light) using the CyberStrike logo
• Added MCP ecosystem showcase — hackbrowser-mcp, cloud-audit-mcp, github-security-mcp, cve-mcp, osint-mcp
• Added GitHub issue templates (bug report, feature request, security tool request)
• Added PR template with security impact section
• Added SECURITY.md with threat model and disclosure policy
• Added CONTRIBUTING.md with MCP ecosystem and community links
• Fixed Config discovery: findUp correctly locates .cyberstrike/ config files
• Fixed Lazy tool registry refresh on ToolListChanged notification
• Fixed Bus.subscribe deferred to init() to avoid Instance context error
• Fixed Bolt auth error detection and user feedback
• Fixed opentui border rendering bug in prompt input
• Fixed @opentui/core updated 0.1.79 → 0.1.87
• Fixed SVG <filter> elements removed for GitHub rendering compatibility
• Fixed Discord invite link fixed across all files (discord.gg/snunAaHf6U)
• Fixed LSP section hidden from sidebar (not applicable to security agents)
• Added Bolt delete keybind changed from d to Ctrl+D to avoid conflicts with search input
• Added 13+ specialized security agents — cyberstrike (primary), web-application, mobile-application, cloud-security, internal-network
• Added 8 proxy testing agents — IDOR, Authorization Bypass, Mass Assignment, Injection, Authentication, Business Logic, SSRF, File Attacks
• Added 120+ OWASP WSTG test cases built into agent methodology
• Added Common vulnerability testing prompt shared across all proxy agents
• Added Vulnerability reporting dialog in TUI
• Added Web proxy context dialog and request detail tools
• Added Vulnerability PoC and business impact DB migrations
• Added Default agent renamed from build to cyberstrike
• Added All specialty agents visible in tab agent list
• Added Removed plan mode tools and workflow
• Fixed Crash in error handler (missing ResolveMessage import)
• Fixed Stale “build” agent references in config schema
• Fixed Debug/placeholder fields removed from package.json
• Added Initial public release of CyberStrike
• Added Fork of opencode with offensive security focus
• Added Claude Code CLI/API provider integration
• Added Cloud security agent
• Added Chunked context compaction and pre-compaction memory flush
• Added Browser tool for default agents
• Added MCP browser server for Claude CLI tool support
• Added ASCII logo with theme colors
• Fixed Rebrand: OpenCode → CyberStrike across terminal title, sidebar, all references
• Fixed Claude CLI timeout increased to 15 minutes
• Fixed Session-id continuity for multi-turn tool calling
• Fixed Infinite loop prevention in multi-step tool calling
• Fixed Empty endpoint migration placeholder

v1.1.14

• Added /methodology dialog in TUI to inspect current methodology state
• Added Sidebar digest showing real-time methodology progress
• Added Intelligent orchestrator delegation based on methodology context
• Added 20 agent bug fixes in a single pass — context handling, prompt formatting, tool routing
• Added Cumulative token usage tracking across main agent + all sub-agents displayed in TUI
• Added Sub-agent prompt payloads capped to prevent context overflow
• Added Anthropic Claude — Pro/Max subscription support via OAuth (use your included API quota directly)
• Added DeepSeek V4 — Pro and Flash variants with thinking mode and reasoning effort control
• Added OpenAI GPT-5 — reasoning summary support for non-chat models
• Added GitHub Copilot — Enterprise support with responses API routing
• Added Local LLMs — Ollama, LM Studio with CLI management and UI improvements
• Added Major: Anthropic, OpenAI, Google (Gemini/Vertex), Azure, Amazon Bedrock, Mistral, Cohere, xAI (Grok)
• Added Performance: Groq, Cerebras, Fireworks AI, Together AI, DeepInfra
• Added Routing: OpenRouter, Requesty, Helicone, Vercel AI
• Added Chinese providers: DeepSeek, Alibaba (Qwen), Tencent, Zhipu AI, Kimi, Minimax, Baichuan, SiliconFlow, StepFun
• Added Cloud: Cloudflare Workers AI, DigitalOcean, Scaleway, OVHCloud, Vultr, Databricks, Snowflake Cortex
• Added New additions: Venice, Nova, Poolside, Trinity, GitLab AI Gateway, SAP AI Core, and 50+ more
• Added Kubernetes v1.11.1, v1.12.0, v2.0.0
• Added Docker v1.7.0, v1.8.0
• Added Apache HTTP Server 2.2 (v3.6.0) and 2.4 (v2.3.0)
• Added Apache Cassandra 3.11, 4.0, 4.1, 5.0
• Added Apache Tomcat 10.1
• Added Ubuntu 18.04, 20.04
• Added Azure Foundations v5.0.0, Compute v2.0.0, Database v2.0.0, Storage v1.0.0
• Added Google Cloud Platform Foundation v4.0.0, Container-Optimized OS v1.2.0, Workspace v1.3.0
• Added MITRE ATT&CK upgraded with Atomic Red Team mappings
• Improved Auth failures now surface immediately instead of producing silent empty crawls
• Improved Subscription parity between worker and main process
• Improved Multi-credential headless mode fixes
• Improved Playwright version pinned exactly in postinstall (no more caret-range drift)
• Improved Agent init: cold-start eliminated with lightweight directory scan
• Improved First message: 20s → <1s via lazy skill loading
• Improved Log growth: 11GB/3min → <100MB (permission evaluate moved to DEBUG)
• Improved Test suite: 64 pre-existing failures resolved
• Improved Session context: tool-error strings truncated before storage
• Improved Permission system: DeniedError messages capped instead of serializing full ruleset
• Added Vulnerability deduplication by endpoint + attack vector
• Added Prompt hardening to block 403/public-endpoint false positives
• Added Session context integration for smarter assessment

v1.1.13

• Added Vulnerability deduplication by endpoint + attack vector (eliminates duplicate findings)
• Added Prompt hardening to block 403/public-endpoint false positives
• Added Session context integration for smarter vulnerability assessment
• Added Added endpoint and attack_vector fields to vulnerability tracking
• Added New tools: web-get-vulnerabilities and web-get-vuln-detail
• Added Unit tests for dedup, tools, and session context
• Improved Copy button in vulnerability detail panel header
• Improved Text selection enabled in vulnerability detail panel
• Added Added HackBrowser section to README
• Added Expanded proxy tester documentation

v1.1.12

• Security Autonomous browser-based vulnerability scanner with Playwright integration
• Security Multi-credential support with manual-login and automated modes
• Security Live telemetry panel injected into target pages
• Security Intelligence Layer: priority scoring, journey awareness, out-of-scope filtering
• Security TUI integration: launch dialog, sidebar status, LLM cost tracking
• Security /hackbrowser slash command for quick launch
• Security Headless mode support
• Security Stop mechanism (/hackbrowser-stop)
• Security Hackbrowser subprocess isolation from main binary
• Improved Agent initialization: 4.3s → ~500ms (lightweight directory scan with Skill.dirsOnly())
• Improved First message response: 19.9s → <1s (lazy skill loading in SkillTool)
• Improved Log file growth: 11GB/3min → <100MB (permission evaluate at DEBUG level + filtered invalid rules)
• Improved Startup warnings: 15,214 duplicate skill warnings silenced (moved to DEBUG)
• Improved Skill permission checks: Eliminated repeated permission checks and index rebuilds
• Improved Reworked agent prompts for web-application, cloud-security, internal-network, and mobile-application testers
• Improved Statically injected WSTG skills into vulnerability tester agents
• Improved Orchestrator web-proxy-agent prompt improvements
• Added 4-tier URL path normalization pipeline
• Added Ingest queue with pause/resume support (/qpause, /qresume)
• Added Strict ingest isolation (Katman 3 hardening)
• Added Ingest context management with excludeHistory and IngestSummary
• Fixed Fixed undefined permission rule errors causing TypeError
• Fixed Fixed TypeScript agent type mismatch in SkillTool execute context
• Fixed Fixed Prettier breaking skill signatures (added .cyberstrike/skill/** to .prettierignore)
• Fixed Fixed massive log spam in permission evaluate
• Fixed Fixed Playwright external build + npm dependency issues
• Fixed Fixed hackbrowser dialog launch bug

v1.1.11

• Fixed Fixed skill loading performance - eliminated 4.3s cold-start delay
• Fixed Fixed first message response time (19.9s → <1s with lazy loading)
• Fixed Fixed permission log spam creating 11GB log files
• Fixed Fixed duplicate skill warnings (15,214 → DEBUG level)
• Fixed Fixed undefined permission rule errors
• Fixed Fixed TypeScript agent type errors in skill tool
• Fixed Protected SKILL.md files from Prettier formatting (added to .prettierignore)
• Improved Agent initialization: 4.3s → ~500ms (lightweight directory scan)
• Improved First message: 19.9s → <1s (lazy skill loading)
• Improved Log files: 11GB/3min → <100MB (debug level + filtered logs)
• Improved Startup: Silenced 15,214 duplicate skill warnings

v1.1.10

• Added MITRE ATT&CK Integration: 691 enterprise techniques with Atomic Red Team tests
• Added 332 techniques now include 2,000+ copy-paste ready test commands
• Added Coverage: Credential Access, Defense Evasion, Discovery, Execution, Persistence, etc.
• Added Example: T1558.003 Kerberoasting (7 Atomic tests), T1003.001 LSASS Memory (5 tests)
• Added CIS Benchmarks: 1,500+ hardening and compliance checks
• Added Cloud Providers: AWS, Azure, GCP, Google Workspace
• Added Server Software: Apache HTTP Server 2.2/2.4, Apache Cassandra 3.11/4.0/4.1/5.0, Apache Tomcat 7/8/9/10
• Added Container/Orchestration: Docker v1.6/1.7/1.8, Kubernetes
• Added Operating Systems: Ubuntu 18.04 LTS, Ubuntu 20.04 LTS
• Added OWASP WSTG: 125 web application security testing skills
• Added NIST: Security controls and frameworks
• Added Lazy Loading: Skills loaded on-demand, zero context pollution
• Added Relevance-based scoring algorithm
• Added Pagination with feedback
• Added Search by: keyword, tech_stack, CWE ID, category, tags
• Added 7,633 skills indexed in-memory
• Added CIS skill recommendations integration
• Added Automated compliance checks

v1.1.9

• Added cyberstrike web — Launches CyberStrike with built-in Web UI on localhost:4096
• Added cyberstrike serve — Headless server mode for remote access via Cloudflare Tunnel
• Added app.cyberstrike.io — Connect to any CyberStrike instance using the hosted Hub UI. Enter your CF tunnel URL and password to connect from anywhere.
• Added Hub Connect Screen — First-time users on app.cyberstrike.io see a connect screen to choose localhost or enter a remote tunnel URL
• Added Password Auth — Set CYBERSTRIKE_SERVER_PASSWORD for remote access. Unicode-safe Basic auth with automatic loopback bypass for local use.
• Added Plugin tools (subfinder, nmap, nuclei, etc.) each appear as separate entries with tool counts
• Added User-configured MCP servers (dnsx, sqlmap, katana, etc.) added via bolt.config.json are also visible
• Added Each Bolt-sourced tool shows a “bolt” badge identifying the source server
• Security osint (37 tools) — OSINT reconnaissance and intelligence gathering
• Security cve (23 tools) — CVE/NVD/EPSS/KEV/GHSA/OSV vulnerability intelligence
• Security github-security (39 tools) — GitHub security scanning with 45 checks
• Improved CORS and auth fixes for CF tunnel remote access
• Improved Retry provider API calls on 5xx server errors
• Improved Bump minimatch and dompurify for security patches
• Improved MCP/Bolt config persisted to global scope
• Improved Web UI bundled in npm package, auto-installed to ~/.cyberstrike/web/

v1.1.5

• Improved Theme hot-reload support
• Improved Select first item automatically when filtering lists
• Fixed Clear palette cache on theme change
• Fixed Encode non-ASCII directory paths in HTTP headers

v1.1.4

• Improved Auto-fallback to available port when default port (4096) is busy
• Improved Add --beta flag to install script for beta channel installs
• Fixed Update @opentui/core and @opentui/solid to 0.1.88
• Fixed Fix scoped package names in uninstall script
• Fixed Correct npm install commands and domain references on website

v1.1.3

• Security Add offensive security agent prompts for all specialized agents
• Security Harden publish script
• Fixed Add billing header for OAT token auth on Sonnet/Opus models — fixes 401 errors with certain API configurations

v1.1.2

• Added Rename opencode → cyberstrike in bin launcher script
• Added Publish only on manual workflow trigger, not on every push
• Added Remove AUR and Homebrew from publish pipeline

v1.1.1

• Added Vulnerability reporting and request normalization models
• Added Web security testing infrastructure (credentials, roles, objects, functions)
• Added report_vulnerability tool with severity and impact tracking
• Added Web proxy agent tools for endpoint analysis and credential management
• Security Vulnerability testing and proxy analysis agent prompts
• Security Extended agent definitions with proxy agents and vulnerability testers
• Security Request context prepending for vulnerability agents
• Added Vulnerability, request, and web context API endpoints
• Added Vulnerability and web context views in TUI session sidebar
• Added Sync context handlers for vulnerability and web security data
• Added Integrated vulnerability reporting into CLI commands
• Added Migrated all NPM packages to @cyberstrike-io/ scope
• Added Redesigned SVG banners, reordered README headers
• Added Added NPM keywords for discoverability
• Fixed Default server port to 4096
• Fixed Fixed migration folder timestamps and loader validation

v0.1.0

• Added 13+ specialized security agents (web, mobile, cloud, network, proxy testers)
• Added 120+ OWASP WSTG test cases
• Added 15+ LLM provider support (OpenAI, Anthropic, Google, AWS Bedrock, Azure, and more)
• Added TUI + Web interfaces
• Added Plugin system for custom agents and tools

Version Format

Cyberstrike follows Semantic Versioning:

Badge Legend

• Added New feature or capability
• Fixed Bug fix
• Improved Enhancement to existing feature
• Security Security-related update
• Breaking Breaking change
• Deprecated Feature marked for removal

Pre-release Versions