Every program here is a benign reproduction of one detection technique. No
malware, no packed payloads, no real-world targeting. Each one triggers
exactly one or two wraith rules and then exits cleanly (/bin/true,
sleep, and a few bytes written into a scratch buffer are the extent of the
"payload"). They exist so a reviewer, or you, can watch wraith fire without
needing an actual sample.
Run wraith in one terminal:
sudo ./wraithThen, in another terminal:
make -C scripts/simulate all # builds the two small C helpers
./scripts/simulate/fileless_exec.sh
./scripts/simulate/preload_hijack.sh
./scripts/simulate/tmpfs_exec.sh
./scripts/simulate/proc_injection.sh| Script | Rule(s) triggered | Technique it stands in for |
|---|---|---|
fileless_exec.sh |
fileless_exec (+ memfd_create with -info) |
Copying /bin/true into a memfd and fexecve-ing it, the same primitive droppers use to run a payload that never touches disk. |
preload_hijack.sh |
preload_hijack |
Forcing a locally-built, no-op shared object into /bin/true via LD_PRELOAD. |
tmpfs_exec.sh |
tmpfs_exec |
Copying /bin/true into /tmp and executing it from there. |
proc_injection.sh |
ptrace_inject, proc_mem_write |
A parent forks a child, PTRACE_ATTACH/PTRACE_DETACHes it, then overwrites a buffer in its address space with process_vm_writev: the two userland primitives behind Linux process injection, used here to write a harmless string instead of shellcode. |
proc_injection only attaches to its own child, so it works even under the
default (ptrace_scope=1) Yama restriction. No capabilities beyond what
your shell already has are required for any of these.