Penetration tester focused on offensive security: AWS/cloud, Active Directory, web, API and AI security. GIAC certified.
I build small, sharp open-source tools for security work. Writing at muratbekgi.com.
| Project | What it does |
|---|---|
| agentcore-iam-audit | Finds Amazon Bedrock AgentCore IAM paths: agents reaching other agents' memory, runtimes and code interpreters, and code-interpreter privilege escalation. |
| mcp-sdk-audit | Finds MCP SDK versions with known advisories, and the code that stays exposed after you upgrade. Zero deps, SARIF, GitHub Action. |
| semgrep-rules-mcp | Semgrep rules for MCP servers: tool poisoning, shell injection from tool arguments, and the SDK settings 2026's advisories made mandatory. |
| nuclei-templates-mcp | Nuclei templates for exposed MCP servers: unauthenticated access, tools anyone can call, CVE-2026-33252 cross-site requests, DNS rebinding. |
| mcp-issuer-check | Shows which OAuth server your MCP servers send your login to, catches issuer impersonation, and alerts when it changes. |
| Project | What it does |
|---|---|
| turkmen-translit | Transliterate Turkmen between the legacy Cyrillic and the current official Latin alphabet — correctly, including the letters other tools get wrong (Ж→ž vs Җ→j, Й→ý, Ң→ň). Zero-dependency Python + JS library + tr-tm CLI. On PyPI and npm. |
| turkmen-num2words | Spell out integers in Turkmen: 1234 → müň iki ýüz otuz dört. Grounded in primary sources and verified against a published grammar's worked example. Zero-dependency Python + JS library + tm-num CLI. |
- a Turkmen transliteration skill and a Turkmen security glossary
- skills-lock: hash-pinning and change alerts for installed Claude/Codex agent skills
- AWS trust-policy drift detection (
UpdateAssumeRolePolicypersistence)
All tools are for authorized testing and defense only.


