Latest
v1.1.16
CyberStrike agents don't just call AI tools, they execute real commands on target systems.
Hook Architecture — Native Post-Exploitation Tools
CyberStrike agents don’t just call AI tools — they execute real commands on target systems. Hooks are native TypeScript post-exploitation programs that generate and run platform-specific commands (PowerShell, cmd, bash, Python, etc.) on compromised machines. Each hook is a single tool call (e.g. winhook credential_dump) that dispatches to a handler producing OS-native commands with fallback chains when primary tools are unavailable.
Why hooks instead of plain tool calls:
- Native execution — runs actual system commands on the target, not API calls
- Fallback chains — if PowerShell is blocked, falls back to cmd.exe/reg.exe/wmic; if bash is unavailable, tries sh/python3/perl
- Structured output — every finding includes CWE mappings, severity, and evidence for automated reporting
- OPSEC built-in — temp files use TMPDIR with try/finally cleanup, forensic secrets are masked in output, credential values are never exposed in findings
This release massively expands the hook arsenal across all platforms and adds new security domains.
Added
- linuxhook — 120 native TypeScript post-exploitation programs across 9 categories: recon, credentials, privesc, persistence, lateral movement, evasion, exfiltration, network attacks, IPv6. Uses bash/sh/python3/perl fallback chain for maximum target compatibility
- llmhook — 23 LLM security testing programs covering OWASP LLM Top 10: prompt injection, jailbreak (15 techniques), training data extraction, hallucination testing, PII/RBAC bypass, multi-turn attacks, indirect injection, evasion (encoding bypass, guardrail detection, token exhaustion), toxicity/bias testing. 130+ probes/payloads with
--format openai|anthropic|genericsupport - ci-audit — 8 read-only CI/CD pipeline security programs for GitHub Actions, GitLab CI, Jenkins, CircleCI
- cloud_audit — native cloud security audit tool for
internal-networkandcloud-securityagents - gcphook — modularized and expanded to 33 GCP post-exploitation programs (from 13)
- LLM security agent (SENTINEL) — proxy-tester-llm tier with methodology engine integration, hackbrowser + generate_report permissions, and llmhook tool registration
- Uninstall wizard — interactive cleanup with running-instance detection and preuninstall hook for package managers
- Skills:
linux-postexploit(kill chain + MITRE ATT&CK),ci-assessment - cve-mcp dynamic CTA — all OS hooks now guide agents to query
cve-mcpfor discovered services, enabling automatic CVE lookup during post-exploitation
Changed — Major Hook Expansions
- winhook — modular refactor: monolithic file → 16 modules, 12 → 154 programs. Added cmd.exe/reg.exe/wmic fallback for restricted environments where PowerShell is blocked
- machook — modular refactor: monolithic file → 10 modules, 12 → 46 programs. New credential, lateral movement, and evasion handlers
- awshook — expanded from 10 to 92 programs across 7 categories (recon, credentials, privesc, persistence, lateral movement, evasion, exfiltration)
- azurehook — expanded from 8 to 154 programs across 13 categories including CIS Azure Benchmark compliance (18 handlers), M365, Entra ID identity attacks, and impact/disruption handlers
- containerhook — expanded from 7 to 20 Docker container security programs
- kubehook — expanded from 7 to 20 Kubernetes security programs
- iachook — expanded from 8 to 15 Terraform/IaC audit programs
- cipipe (azurehook, kubehook) — converted from Python to native TypeScript
macos-postexploitskill — updated from 12 to 46 programs
Fixed
- Hackbrowser: shell quoting error in Chromium path detection (nested double-quote collision)
- Hackbrowser: Esc now stops the crawl subprocess — previously continued running in background after session cancel
- Build: bundled migration crash on first-time setup (Drizzle ORM 1.0 missing
namefield) - Provider: Vertex AI autoload with
GOOGLE_APPLICATION_CREDENTIALSfor both Gemini and Claude, EC2 Instance Role auth for Bedrock (#58), full AWS credential chain with IMDS fallback - Hooks (all platforms): duplicate checkId resolution across all modules, TMPDIR usage for temp files with try/finally cleanup, forensic secret masking in output, user-supplied regex validation, JSON.parse guards on external input, handler dispatch error handling
- Session: content-filter finish reason surfaced as visible error instead of silent exit
- Bash tool: binary output detection and stripping to prevent terminal corruption
- DB: idempotent migration CREATE statements
Security
- fast-xml-parser 5.9.3 → 5.10.1 (DoS via XML parsing)
- dompurify 3.4.11 → 3.4.13 (XSS via hook removal)
Community Contributions
- @csmcguire — EC2 Instance Role auth for Amazon Bedrock (#58)
- @Nislive — LLM security agent (#33)
- @Serhatcck — llmhook refusal/canary detection, llm-security agent tier restructure