zellij-session-snapshot

Warn

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script scripts/conwrap.ps1 uses the Add-Type PowerShell command to define and compile C# code at runtime. This code interfaces directly with Win32 APIs (CreateFileW, CreateProcessW) to manipulate process handles and execution.
  • [DYNAMIC_EXECUTION]: Both scripts/snapshot.py and scripts/snapshot_windows.py utilize the ctypes library for Foreign Function Interface (FFI) to load system libraries (libc.dylib, kernel32.dll, ntdll.dll) and perform low-level operations like reading process memory and system-level process information.
  • [COMMAND_EXECUTION]: The skill frequently executes shell commands via subprocess.run, including zellij actions and PowerShell scripts. On Windows, it uses WMI (Win32_Process.Create) to spawn detached background processes, which is a powerful administrative capability.
  • [DATA_EXPOSURE]: The skill performs deep introspection into the memory space and environment of other running processes to identify Claude Code sessions. On macOS, it uses sysctl KERN_PROCARGS2, and on Windows, it uses ReadProcessMemory to extract the environment block from the Process Environment Block (PEB).
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The skill ingests data from local tool outputs (zellij action list-panes) and Claude's internal session files (~/.claude/sessions/*.json).
  • Boundary markers: The skill uses KDL-specific string escaping when generating layout files to be consumed by Zellij.
  • Capability inventory: The skill has the ability to write arbitrary commands into Zellij layout files and spawn processes via WMI.
  • Sanitization: Basic string escaping (kdl_str) is applied to arguments and paths processed by the skill.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 11, 2026, 04:20 AM