rhaist/marq / skills/malware/malware-scripts
malware-scripts
FlaggedGuides statically triaging and deobfuscating script-based lures (LNK, HTA, JS/JScript, VBS/VBE, PowerShell, batch, HTML smuggling, ClickFix-style paste-and-run pages) to recover the next-stage command and IOCs without ever executing them. Use when the user has a suspicious script, shortcut or HTML attachment, an obfuscated PowerShell one-liner, or a fake CAPTCHA page that asks to paste a command.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.0 KB |
SKILL.md
View on GitHub---
name: malware-scripts
description: Guides statically triaging and deobfuscating script-based lures (LNK, HTA, JS/JScript, VBS/VBE, PowerShell, batch, HTML smuggling, ClickFix-style paste-and-run pages) to recover the next-stage command and IOCs without ever executing them. Use when the user has a suspicious script, shortcut or HTML attachment, an obfuscated PowerShell one-liner, or a fake CAPTCHA page that asks to paste a command.
license: Apache-2.0
metadata:
verified: "2026-11-17"
---
# Script & lure analysis (static)
**Never execute.** Don't run the sample, and don't run its decoder either, not even with
`eval` swapped for a print: that is still running attacker code. marq ships no
`pwsh`/`node`/`wscript`, which is correct. Recover each layer by re-implementing the
decode step yourself. Stage under `/work/<case>/`; keep your own decode scripts in a
different dir and run them with `python3 -I`. Malicious `.doc(m)`/`.xls(m)`/OneNote →
`load_skill malware-maldoc`; a PE/ELF that falls out → `load_skill malware-triage`.
## Identify & unpack the container
- Hash first (`sha256sum`) and check reputation (`load_skill indicator-enrichment`).
- Type by content, not extension: `exif_metadata` reports File Type / MIME Type. `read_file`
with a small `max_bytes` shows the head of a text script.
- ZIP: `unzip -l`, then `unzip -d /work/<case>/x` (unpacking runs nothing).
ISO/IMG/VHD(X)/7z/RAR: unpack with 7-Zip and stage the contents.
## Per lure type
- **LNK**: `exif_metadata` gives the target path, **command-line arguments**, working dir, icon
file (a PDF icon on a `.lnk` = lure) and Machine ID (the builder's hostname, a pivot).
Arguments padded with whitespace to hide past the Properties box (CVE-2025-9491); Windows
changed that dialog in 2025, but it's a UI-only fix. Read the raw args, not the GUI.
verify: https://0patch.com/blog/microsoft-silently-patched-cve-2025-9491-we-think-our-patch-provides-more-securit
Payloads appended after the LNK structure: `rabin2 -zz` (raw + UTF-16 strings).
- **HTA**: HTML run by `mshta.exe` at full trust, outside the browser sandbox. `read_file`; look
for `<script language="VBScript">`, `CreateObject("WScript.Shell")`, `.Run`, `window.close`.
- **JS/JScript (WSH)**: a double-clicked `.js` runs in `wscript.exe`, not a browser. Sinks and
primitives: `ActiveXObject`, `WScript.Shell`, `MSXML2.XMLHTTP`, `ADODB.Stream` (`SaveToFile`),
`eval`/`Function()`. Obfuscation: a string array + rotate/index function (obfuscator.io style),
`fromCharCode`, reversed strings, megabytes of junk padding. Lift the array and decoder logic
into python3. Emulation (box-js) belongs off-box, in a sandbox.
- **VBS/VBE**: `olevba` (oletools) reads plain VBScript text too (auto-exec, suspicious keywords, `--deobf`).
`.vbe` (`#@~^` header) is Script-Encoder output: decode off-box (Didier Stevens' `decode-vbe.py`)
or re-implement it. VBScript is still on by default in Windows 11; it's planned to stop being
on by default around 2027. verify: https://techcommunity.microsoft.com/blog/windows-itpro-blog/vbscript-deprecation-timelines-and-next-steps/4148301
- **PowerShell**: `-e`/`-enc`/`-EncodedCommand` is base64 of UTF-16LE: `base64 -d | iconv -f
UTF-16LE -t UTF-8`. Next layers: `FromBase64String` + `GzipStream`/`DeflateStream` (python3
`gzip`, or `zlib.decompress(d, -15)` for raw deflate), `-bxor` keys, `[char]` arrays + `-join`,
`-f` format reordering, backticks. Sinks: `IEX`, `Invoke-Expression`, `&(...)`, `.Invoke()`.
Strip the sink and decode its argument. `AmsiUtils`/`amsiInitFailed` = AMSI bypass.
- **Batch (.bat/.cmd)**: `set` fragments joined via `%v:~x,y%` substrings, `^` carets, undefined
`%junk%` vars that expand to nothing, a UTF-16 BOM to break editors, `certutil -decode`, and
payloads after `exit /b`/`goto :eof` read back by `findstr`/PowerShell. Resolve the `set`s in python3.
- **HTML smuggling** (T1027.006): an HTML/SVG attachment builds a file in-browser (`Blob`,
`URL.createObjectURL`, `a.download`, `msSaveOrOpenBlob`) from an embedded base64/array blob,
often XOR'd or reversed. Regex the blob out, decode it in python3, write it to `/work` as an
inert file and triage that (often ZIP/ISO → LNK). Credential-phish pages: note the POST target.
- **ClickFix / paste-and-run** (ATT&CK T1204.004, 2025): a fake CAPTCHA/"fix it" page writes a
command to the clipboard (`navigator.clipboard.writeText`, `execCommand('copy')`) and tells the
user to paste it into Win+R, Terminal (Win+X → I), the Explorer address bar ("FileFix") or
macOS Terminal. Typical: `mshta https://…`, `powershell -w h … | iex`, `curl … | bash`, with a
trailing `# I'm not a robot` comment. verify: https://attack.mitre.org/techniques/T1204/004/
Get the page HTML from email/proxy logs or an existing urlscan scan. Don't browse it from an
engagement box. If the command is only fetched at runtime (remote API, blockchain contract),
you need `load_skill malware-dynamic`. IR artefact: pasted Run-dialog commands land in
`HKCU\Software\Microsoft\Windows\CurrentVersion\Explorer\RunMRU`. macOS 26.4+ warns on suspicious
Terminal pastes. verify: https://support.apple.com/en-us/127377
## Deobfuscation discipline
- One layer at a time; save each as `layerN.txt` and hash the final payload.
- The goal is the **next-stage URL/command and dropped filename**, not a pretty script.
- Never fetch the stage-2 from marq: enrich the URL (`load_skill indicator-enrichment`) or send
it to a sandbox. Stuck (environment-keyed decryption, runtime fetch, a decoder too complex to
port)? → `load_skill malware-dynamic`.
## Detect & report
- YARA for lures: text strings with `nocase`, `wide` for UTF-16 (LNK args); anchor on the
obfuscator's unique function/variable names plus a sink. Test with `yara`; method in
`load_skill malware-yara`.
- `report_finding`: evidence = delivery chain (container → LNK → `mshta` URL → stage-2), the
defanged deobfuscated command, IOCs, ATT&CK (T1204.002/.004, T1027.006, T1059.001/.003/.005/.007,
T1218.005 Mshta). Pivot IOCs via `load_skill ioc-pivoting`.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |