Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Identifies and unpacks UPX-packed and other packed malware samples to expose the original executable code for static analysis. Covers both standard UPX unpacking and handling modified UPX headers that prevent automated decompression. Activates for requests involving malware unpacking, UPX decompression, packer removal, or preparing packed samples for analysis.
.claude/skills/analyzing-packed-malware-with-upx-unpacker/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | — | — |
| case-17 | ✗→✓ | ▲ Improved | — | — |
| case-18 | ✗→✓ | ▲ Improved | — | — |
| case-03 | ✗→✓ | ▲ Improved | — | — |
| case-16 | ✗→✓ | ▲ Improved | — | — |
Do not use when dealing with custom packers, VM-based protectors (Themida, VMProtect), or samples where dynamic unpacking via debugging is more appropriate.
apt install upx-ucl or download from https://upx.github.io/)pefile library for manual header repairDetermine if the sample is packed and identify the packer:
bash# Check with Detect It Easy diec suspect.exe # Check with UPX (test without unpacking) upx -t suspect.exe # Python-based entropy and packer detection python3 << 'PYEOF' import pefile import math pe = pefile.PE("suspect.exe") print("Section Analysis:") for section in pe.sections: name = section.Name.decode().rstrip('\x00') entropy = section.get_entropy() raw = section.SizeOfRawData virtual = section.Misc_VirtualSize print(f" {name:8s} Entropy: {entropy:.2f} Raw: {raw:>8} Virtual: {virtual:>8}") # Check for UPX section names section_names = [s.Name.decode().rstrip('\x00') for s in pe.sections] if 'UPX0' in section_names or 'UPX1' in section_names: print("\n[!] UPX section names detected") elif '.upx' in [s.lower() for s in section_names]: print("\n[!] UPX variant section names detected") # Check import count (packed binaries have very few) if hasattr(pe, 'DIRECTORY_ENTRY_IMPORT'): total_imports = sum(len(e.imports) for e in pe.DIRECTORY_ENTRY_IMPORT) print(f"\nTotal imports: {total_imports}") if total_imports < 10: print("[!] Very few imports - likely packed") else: print("\n[!] No import directory - heavily packed") PYEOF
Try the built-in UPX decompression:
bash# Standard UPX decompress upx -d suspect.exe -o unpacked.exe # If UPX fails with "not packed by UPX" error, the headers may be modified # Verbose output for debugging upx -d suspect.exe -o unpacked.exe -v 2>&1 # Verify the unpacked file file unpacked.exe diec unpacked.exe
If standard decompression fails, repair tampered magic bytes:
python# Repair modified UPX headers import struct with open("suspect.exe", "rb") as f: data = bytearray(f.read()) # UPX magic bytes: "UPX!" (0x55505821) # Malware authors commonly modify these to prevent automatic unpacking # Search for modified UPX signatures upx_magic = b"UPX!" modified_patterns = [b"UPX0", b"UPX\x00", b"\x00PX!", b"UPx!"] # Find and restore section names pe_offset = struct.unpack_from("<I", data, 0x3C)[0] num_sections = struct.unpack_from("<H", data, pe_offset + 6)[0] section_table_offset = pe_offset + 0x18 + struct.unpack_from("<H", data, pe_offset + 0x14)[0] print(f"PE offset: 0x{pe_offset:X}") print(f"Number of sections: {num_sections}") print(f"Section table offset: 0x{section_table_offset:X}") for i in range(num_sections): offset = section_table_offset + (i * 40) name = data[offset:offset+8] print(f"Section {i}: {name}") # Restore UPX magic bytes in the binary # Search for the UPX header signature location (typically near the end of packed data) for i in range(len(data) - 4): if data[i:i+3] == b"UPX" and data[i+3] != ord("!"): print(f"Found modified UPX magic at offset 0x{i:X}: {data[i:i+4]}") data[i:i+4] = b"UPX!" print(f"Restored to: UPX!") # Also restore section names if modified for i in range(num_sections): offset = section_table_offset + (i * 40) name = data[offset:offset+8].rstrip(b'\x00') if name in [b"UPX0", b"UPX1", b"UPX2"]: continue # Already correct # Check for common modifications if name.startswith(b"UP") or name.startswith(b"ux"): original = f"UPX{i}".encode().ljust(8, b'\x00') data[offset:offset+8] = original print(f"Restored section name at 0x{offset:X} to {original}") with open("suspect_fixed.exe", "wb") as f: f.write(data) print("\nFixed file written. Retry: upx -d suspect_fixed.exe -o unpacked.exe")
When automated unpacking fails entirely, use dynamic unpacking:
Manual UPX Unpacking with x64dbg:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1. Load packed sample in x64dbg
2. Run to the entry point (system breakpoint then F9)
3. UPX unpacking stub pattern:
a. PUSHAD (saves all registers)
b. Decompression loop (processes packed sections)
c. Resolves imports (LoadLibrary/GetProcAddress calls)
d. POPAD (restores registers)
e. JMP to OEP (original entry point)
4. Set hardware breakpoint on ESP after PUSHAD:
- After PUSHAD, right-click ESP in registers -> Follow in Dump
- Set hardware breakpoint on access at [ESP] address
- Run (F9) - breaks at POPAD before JMP to OEP
5. Step forward (F7/F8) until you reach the JMP to OEP
6. At OEP: Use Scylla plugin to dump and fix imports:
- Plugins -> Scylla -> OEP = current EIP
- Click "IAT Autosearch" -> "Get Imports"
- Click "Dump" to save unpacked binary
- Click "Fix Dump" to repair import tableVerify the unpacked sample is valid and complete:
bash# Verify unpacked PE is valid python3 << 'PYEOF' import pefile pe = pefile.PE("unpacked.exe") # Check sections are normal print("Unpacked Section Analysis:") for section in pe.sections: name = section.Name.decode().rstrip('\x00') entropy = section.get_entropy() print(f" {name:8s} Entropy: {entropy:.2f}") # Verify imports are resolved print(f"\nImport count:") if hasattr(pe, 'DIRECTORY_ENTRY_IMPORT'): for entry in pe.DIRECTORY_ENTRY_IMPORT: dll = entry.dll.decode() count = len(entry.imports) print(f" {dll}: {count} functions") total = sum(len(e.imports) for e in pe.DIRECTORY_ENTRY_IMPORT) print(f" Total: {total} imports") # Compare file sizes import os packed_size = os.path.getsize("suspect.exe") unpacked_size = os.path.getsize("unpacked.exe") print(f"\nPacked: {packed_size:>10} bytes") print(f"Unpacked: {unpacked_size:>10} bytes") print(f"Ratio: {unpacked_size/packed_size:.1f}x") PYEOF
| Term | Definition | |------|------------| | Packing | Compressing or encrypting executable code to reduce file size and hinder static analysis; the binary contains an unpacking stub that restores code at runtime | | UPX | Ultimate Packer for eXecutables; open-source executable packer commonly abused by malware authors because it is free and effective | | Original Entry Point (OEP) | The real starting address of the malware code before packing; the unpacking stub decompresses code then jumps to the OEP | | Import Reconstruction | Process of rebuilding the import address table after dumping an unpacked process from memory using tools like Scylla or ImpRec | | PUSHAD/POPAD | x86 instructions that save/restore all general-purpose registers; UPX uses this pattern to preserve register state during unpacking | | Section Entropy | Randomness measure of PE section data; packed sections show entropy > 7.0 while normal code sections average 5.0-6.5 | | Magic Bytes | Signature bytes within a file identifying its format; UPX uses "UPX!" which malware authors modify to prevent automated decompression |
Context: A malware sample is identified as UPX-packed by section names (UPX0, UPX1) but upx -d fails with "CantUnpackException: header corrupted". The malware author modified the UPX magic bytes to prevent automated decompression.
Approach:
upx -d on the repaired binaryPitfalls:
UNPACKING ANALYSIS REPORT
===========================
Sample: suspect.exe
SHA-256: e3b0c44298fc1c149afbf4c8996fb924...
Packer: UPX 3.96 (modified headers)
PACKED BINARY
Sections: UPX0 (entropy: 0.00) UPX1 (entropy: 7.89) .rsrc (entropy: 3.45)
Imports: 2 (kernel32.dll: LoadLibraryA, GetProcAddress)
File Size: 98,304 bytes
UNPACKING METHOD
Method: Header repair + UPX -d
Header Fix: Restored UPX! magic at offset 0x1F000
Command: upx -d suspect_fixed.exe -o unpacked.exe
Result: SUCCESS
UNPACKED BINARY
Sections: .text (entropy: 6.21) .rdata (entropy: 4.56) .data (entropy: 3.12) .rsrc (entropy: 3.45)
Imports: 147 (kernel32, user32, advapi32, wininet, ws2_32)
File Size: 245,760 bytes (2.5x expansion)
OEP: 0x00401000
VALIDATION
PE Valid: Yes
Imports Resolved: Yes (147 functions across 8 DLLs)
Executable: Yes (runs without crash in sandbox)
NEXT STEPS
- Import unpacked.exe into Ghidra for full disassembly
- Run YARA rules against unpacked binary
- Submit unpacked binary to VirusTotal for improved detection| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-08 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +23 percentage points is the difference between those two pass rates over the 22 comparable cases.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.