DeadPineapple/zombiemutex

Jeffrey Monica

@DeadPineapple

모바일 플랫폼, 백엔드 시스템, 보안 엔지니어링을 다룹니다. 직접 작성한 코드와 테스트, 재현한 오류, 그 문제를 해결한 이유를 기록합니다.

소개 GitHub

현재 만들고 있는 것

ZombieMutex

보안 사례 분석, 실습실과 이 엔지니어링 기록의 개선 작업입니다.

개발 기록 보기

작업실에서

코드, 기록 및 일상

Weekly noteWeek of 2026년 11월 2일

Weekly experiment

Making this page feel less like a changelog and more like a workbench: real code activity, screenshots, small bugs, and the occasional life detour.

What I'm trying

The fiddly bit is knowing what belongs in public. Public commits should link to the actual changes, private builds should stay private, and a photo only goes up if I actually took it. Also trying to keep the whole thing from turning into an airline cockpit.

My current projects

DeadPineapple from the workbench

2026년 11월 9일

Security engineering

Put the proof next to the code

Captured fresh Rust and Go test output and a live browser policy trace, then put each next to the implementation it actually tests.

ELFGuard's six tests and the bounded mutation regression remain local static-analysis evidence. AuthRace's 24 competing HTTP mutations show 120 units in the deliberately vulnerable handler versus 5 with atomic idempotency. For agent security, the browser trace demonstrates a denied tool request; it is not a Linux sandbox attestation.

Rust parser and evidence Go replay reproduction Agent policy trace Open project

DeadPineapple from the workbench

2026년 11월 9일

Architecture notes

The boundary is not just a diagram box

Mapped three actual implementation paths: bytes into the ELF parser, identity through the API mutation ledger, and agent intent into a policy decision.

A box marked 'secure' doesn't help anyone review a system. The useful parts are where untrusted input crosses into authority, what stops it, and what the implementation cannot prove. The Go container runner stays documented as a local reference; no end-to-end Linux kernel isolation claim.

Read the architecture and failure paths Open project

DeadPineapple published a post

2026년 11월 9일

Writing

I built a tiny neural audio codec. Then I listened to the results.

A trained nonlinear encoder, vector-quantized codebook and decoder are interesting. An idealized low bitrate is not the sa…

A trained nonlinear encoder, vector-quantized codebook and decoder are interesting. An idealized low bitrate is not the same thing as good reconstructed speech.

Full post / page

DeadPineapple published a post

2026년 11월 9일

Writing

Audio ML engineering: what the numbers actually mean

I built five small speech and audio implementations and measured their behavior. ASR errors, turn detection, codecs, proso…

I built five small speech and audio implementations and measured their behavior. ASR errors, turn detection, codecs, prosody and classifier ablations each have different ways to mislead you.

Full post / page

DeadPineapple published a post

2026년 11월 9일

Writing

Mobile voice agents: keep the backend in charge

How I connected Swift and Kotlin clients to agent workflows, built Core ML and TensorFlow Lite integration boundaries, and…

How I connected Swift and Kotlin clients to agent workflows, built Core ML and TensorFlow Lite integration boundaries, and kept authorization on the backend.

Full post / page

DeadPineapple published a post

2026년 11월 9일

Writing

Voice agents are distributed systems with microphones

Building an agent that talks is the easy demo. The harder work is authenticated tools, retries, evaluation, telemetry, CRM…

Building an agent that talks is the easy demo. The harder work is authenticated tools, retries, evaluation, telemetry, CRM synchronization and the people who have to operate it.

Full post / page

DeadPineapple published a post

2026년 11월 7일

Writing

Retry storms: when resilience multiplies a failure

Retry logic looks harmless in isolation. When a gateway, application service and SDK each apply their own retry policy, a…

Retry logic looks harmless in isolation. When a gateway, application service and SDK each apply their own retry policy, a single failed operation can multiply into a dependency outage. That interaction is the starting point for my retry-storm analysis.

Full post / page

DeadPineapple from the workbench

2026년 11월 7일

Security lab

A lab that lets you break the assumptions

The device-trust lab is a lot more useful when you can flip the inputs and watch the decision change.

A real capture of the live lab below. Try a replay, a forged integrity signal, or an unauthorized object request. The server is still the one that has to make the authorization decision. The client can be very persuasive; it doesn't get a vote.