Independent security research · Bug bounty hunting · Systems programming
I investigate trust boundaries in developer tools and build systems software. My work includes a responsible disclosure recognized by Anthropic, virtualization research, pre-boot authentication, and reproducible security evaluation.
Independent bug bounty research · Developer-tool security · May 2026
Investigated how repository-provided hooks and MCP configuration interact with workspace trust and noninteractive execution in Anthropic’s coding-agent CLI.
- Prepared reproduction cases, permission-mode comparisons, and startup timing evidence.
- Submitted the research and followed up with Anthropic’s security team to examine alternative explanations.
- Outcome: Anthropic recognized the submission as informative and said it aided the discovery of other vulnerabilities.
Skills: threat modeling, controlled experiments, reproduction design, evidence analysis, and technical disclosure writing.
| Project | Engineering focus | Read more |
|---|---|---|
| Wraith | Linux KVM host, ELF32 guest loading, freestanding C/assembly kernel, and a separate UEFI/AMD SVM research path. | Case study |
| BootGate | UEFI authentication, PBKDF2 password derivation, persistent lockout state, recovery credentials, and optional TPM PCR checks. | Case study |
| Polysec Harness | Python orchestration, bounded parallel investigation, typed evidence, independent verification, and reproduction workflows. | Case study |
| Redteam Harness | Separate evaluation roles, structured attempt records, drift audits, and clearly labeled synthetic replay. | Case study |
| Hostcheck / Leakcheck | Native host-posture adapters and VPN diagnostics with structured findings and explicit inconclusive outcomes. | Hostcheck · Leakcheck |
These repositories are currently private. Their public case studies describe implementation choices, threat boundaries, and validation limits.
- RLBOTSLOT — simulation-based reinforcement learning with object-centric observations, recurrent policy state, PPO, and vectorized rollout infrastructure. Research prototype; current evidence and limitations.
- Gmail Manager — OAuth-backed API integration, batch workflows, and shared Python services behind a CLI and local FastAPI UI.
- BL2 Save Editor — binary-format integration, bit-packed item serialization, per-file locking, atomic replacement, and optional 3D asset visualization.
- flaggle.lofi — deterministic daily state, geospatial/color algorithms, versioned browser persistence, and procedural Web Audio.
- RL Map Manager — PySide6 catalog integration, metadata caching, and content-installation workflows.
- Roblox Multi-Instance — C# / WPF, native resource lifetime, DPAPI-backed storage, MVVM, and DWM window previews.
| Area | Evidence in the projects |
|---|---|
| Systems programming | C, x86 assembly, ELF loading, Linux KVM, UEFI/EDK II, AMD SVM |
| Security engineering | Threat models, authentication policy, host/network diagnostics, evidence-based evaluation |
| Python tooling | Typed schemas, subprocess orchestration, filesystem IPC, CLI/API services, test fixtures |
| Application systems | TypeScript/JavaScript, relational persistence, state machines, browser APIs, C#/.NET |
| Applied ML | PyTorch, PPO, recurrent state, rollout collection, synthetic pipeline validation |
For security, systems, and computer-science engineering opportunities: killianmiller6@gmail.com.
