Skip to content
View Atomics-hub's full-sized avatar
🦏
🦏
6 followers · 2 following

Achievements

Achievement: QuickdrawAchievement: Pull Sharkx3Achievement: YOLO

Achievements

Achievement: QuickdrawAchievement: Pull Sharkx3Achievement: YOLO

Block or report Atomics-hub

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Atomics-hub/README.md

Tom Ryan / A5omic

Security researcher and software engineer focused on memory safety, authorization boundaries, and reproducible proof.

Website · Security ledger · HackerOne · Writing · PGP

Selected security work

Surface Result Public proof
Linux io_uring Found an SQE_MIXED out-of-bounds read. The fix merged upstream and the regression test landed in liburing. Kernel fix · liburing test
V8 / Chrome Found an uninitialized read in Maglev. The V8 team fixed it and listed the work on the Chrome VRP panel. V8 fix · Writeup
trigger.dev Reported a cross-tenant task replay issue, fixed in v4.5.2. GHSA-9fq3-68cw-r7p2
Vaultwarden Independently reported an SSO email verification bypass, published as CVE-2026-47164 and fixed in 1.36.0. Writeup · Advisory
Anthropic One report resolved through HackerOne with a $100 bounty. Technical details are not public. HackerOne profile

How I work

The edge is targeting and proof, not volume. I use focused fuzzers, source review, and target-native test harnesses to find suspicious boundaries. Every promising result gets reproduced in the target's compiled code before it reaches a maintainer. A plausible report is not enough. The bar is evidence that survives a hostile read.

Building

Project What it is
reproassert Generates and sandbox-verifies failing pytest reproduction candidates from GitHub issues.
AgentK A user-space security kernel for AI agents with typed tool boundaries, policy, receipts, and replayable evidence.
flowdown A zero-dependency, O(1) streaming Markdown renderer for AI output.

Elsewhere

I ship apps through Rekishi LLC, write technical and personal essays at tomryan.dev, release music as A5omic, and publish consciousness research on Zenodo.

Contact: overboardapps@gmail.com · GitHub · HackerOne

Pinned Loading

  1. Linux Kernel io_uring: OOB Read via ... Linux Kernel io_uring: OOB Read via SQE_MIXED + sq_array Physical Index Bypass
    1 
    # Linux Kernel io_uring: Out-of-Bounds Read via SQE_MIXED + sq_array Physical Index Bypass
    2 
    
              
    
              
    
                3
                
    **CVE**: Pending (kernel fix and liburing test both merged)
    
              
    
              
    
                4
                
    **Affected**: Linux kernels with `IORING_SETUP_SQE_MIXED` (introduced by commit [1cba30bf9fdd](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=1cba30bf9fdd))
    
              
    
              
    
                5
                
    **Fix**: [c76e0f1d77f8](https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux.git/commit/?id=c76e0f1d77f87e258193c2628253782d5ff414c7) — merged into `io_uring-7.0` on 2026-03-10
    
              
    
          
    
    
    
  
    

    2. 

      

  2. 
  
    
    
    
      
    
        
        V8 Maglev JIT: Uninitialized Specula...  V8 Maglev JIT: Uninitialized SpeculationMode in SaveCallSpeculationScope (CL 7651434)      
    

          
    
              
    
                1
                
    // PoC: Uninitialized SpeculationMode in SaveCallSpeculationScope
    
              
    
              
    
                2
                
    //
    
              
    
              
    
                3
                
    // Triggers uninitialized read of saved_mode_ in the destructor of
    
              
    
              
    
                4
                
    // MaglevGraphBuilder::SaveCallSpeculationScope during Maglev JIT compilation.
    
              
    
              
    
                5
                
    //
    
              
    
          
    
    
    
  
    

    3. 

      

  3. 
  
    
    
    
      
    
        
        CVE-2026-47164 / GHSA-6x5c-84vm-5j56...  CVE-2026-47164 / GHSA-6x5c-84vm-5j56 — Vaultwarden SSO existing-user binding bypasses IdP email_verified check (HIGH 7.7, fixed in 1.36.0). Independent-reporter writeup.      
    

          
    
              
    
                1
                
    # Vaultwarden SSO existing-user binding skips the IdP `email_verified` check (CVE-2026-47164)
    
              
    
              
    
                2
                
    
              
    
              
    
                3
                
    **Advisory:** [GHSA-6x5c-84vm-5j56](https://github.com/dani-garcia/vaultwarden/security/advisories/GHSA-6x5c-84vm-5j56)
    
              
    
              
    
                4
                
    **CVE:** CVE-2026-47164
    
              
    
              
    
                5
                
    **Severity:** HIGH — CVSS 7.7 (`AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:L`)
    
              
    
          
    
    
    
  
    

    4. 

      
  4. 
  
    
    
    
      
    
        
    
            
          slopOS  slopOS          Public
        
    
      
    


      
    
        A TempleOS-inspired OS built from scratch: own bootloader, kernel, and the SlopC language. 640x480, 16 colors, ring 0, version 6.9 forever.
      
    

      
    
          
  
  C


      
    
    
    
  
    

    5. 

      
  5. 
  
    
    
    
      
    
        
    
            
          flowdown  flowdown          Public
        
    
      
    


      
    
        O(1) streaming markdown renderer for the AI era. Zero dependencies. ~4KB gzipped.
      
    

      
    
          
  
  JavaScript


          
            
    

            1
          
      
    
    
    
  
    

    6. 

      
  6. 
  
    
    
    
      
    
        
    
            
          agentk  agentk          Public
        
    
      
    


      
    
        AgentK is a user-space security kernel for AI agents: an MCP/tool boundary with typed syscalls, taint-aware policy, capability receipts, signed evidence, and replayable flight logs.
      
    

      
    
          
  
  Rust


      
    
    
    
  
    

    7.