Day 12 - AutoIt Takes Two (script-analysis)
GM id: script-analysis-day-12-autoit-workflow-pair | lane: script-analysis | fields: 855 | agent: agent-1
Result
| Mode | Result | Flag | Time |
|---|---|---|---|
| Single Solve | CORRECT | Token27a4adffe0955f8bad048a4a55d9a8d9 | 3.1 s |
| Time Trial | CORRECT | Tokend0af1c5dccee188ddd7f93618210726d | 3.2 s |
| Mastery | CORRECT | Tokena395adc89913d7ee97e33d29f7a636be | 4.1 s |
All three modes were solved by the banked lane autosolver (challenges/_lib/autosolve_script.py, run through autosolve.py) at claim time, before any manual work. Nothing was executed from the archive; every answer was derived statically from the fixtures' own bytes.
Challenge
GM prompt: "Two inert AutoIt workflows arrived as a matched set, and each one carries half the story. Correlate them with everything you've proven so far, and decide which pairing is genuine and which is just going through the motions."
Artifact: artifact_0.bin (challenge artifact, not published) - 93,075 byte WinZip-AES zip, 99 members, password infected (stock unzip fails; pyzipper.AESZipFile or 7z x -pinfected).
Member mix: 16 .sh, 14 .py, 12 .js, 10 .vbs, 9 .php, 8 .mjs, 5 .cmd, 5 .lua, 4 .ps1, 4 .bat, 4 .mof, 4 .ts, 4 .au3. Every language layer from Days 1-11 (PowerShell, JScript, Batch, VBScript, Python, Bash, Node .mjs, PHP polyglots, WMI .mof) is still present and still asked about - the lane is cumulative.
Today's new layer: AutoIt .au3
The title and the GM id gave it away, and the four new members confirmed it:
polyglot-autoit-benign-utility.au3
polyglot-autoit-downloader.au3
polyglot-autoit-launcher.au3
polyglot-autoit-staged-chain.au3
44 new question keys appeared, all of the shape polyglot-autoit-<role>-<field> (-role, -attack, -campaign, -url, -file-path, -steps, -stage-count, -adapter, -host-profile, -next-language, -obfuscation, -stage-profile).
Fixture structure
Each .au3 file is the same three-part shape - a self-describing descriptor, exactly as every fixture has been since Day 6:
- A header comment
; HUNTRESS_ANALYSIS_V1:<TAG>-<blob> Global Const $descriptor_<id> = "<same blob>"Func build_plan_<id>() / Return $descriptor_<id> / EndFunc
Decoding the blob yields a pipe-separated k=v descriptor (v=3), which answers every field directly. The only variation is the <TAG> encoding of the blob, one per fixture:
| Tag | Encoding | obfuscation_profile |
|---|---|---|
HEX | plain hex | plain_hex |
B64 | base64 | base64 |
HB64 | hex of base64 (decode hex, then base64) | hex_base64 |
FRAG | hex in dot-separated 12-byte fragments | fragmented_hex |
FRAG is just unhexlify(blob.replace('.', '')); HB64 is b64decode(unhexlify(blob)). All four were already in the adapter's carrier table.
Recovered descriptors (the answers)
| benign-utility | downloader | launcher | staged-chain | |
|---|---|---|---|---|
| role | benign_utility | downloader | launcher | staged_chain |
| verdict | benign | suspicious | suspicious | suspicious |
| attack | none | T1105 | T1059 | T1059 |
| adapter | local_parser | http_client | process_builder | operation_graph |
| host_profile | standalone | standalone | standalone | standalone |
| obfuscation | plain_hex | base64 | hex_base64 | fragmented_hex |
| stage_profile | single_descriptor | two_stage_download_execute | three_stage_decode_write_launch | cross_language_handoff |
| next_language | powershell | jscript | batch | vbscript |
| stage_count | 2 | 2 | 3 | 3 |
| steps | decode_configuration>return_configuration | acquire_remote>dispatch_process | decode_embedded>write_file>dispatch_process | profile_host>decode_configuration>dispatch_language |
| campaign_id | AUTOIT-6fa3e225 | AUTOIT-908af952 | AUTOIT-02a74447 | AUTOIT-1dda6094 |
| url | https://autoit-6fa3e2252a.example.invalid/stage | https://autoit-908af95255.example.invalid/stage | https://autoit-02a744470e.example.invalid/stage | https://autoit-1dda609468.example.invalid/stage |
| file_path | /tmp/huntress-autoit/.6fa3e2252a5414c5.stage | /tmp/huntress-autoit/.908af952550be9dd.stage | /tmp/huntress-autoit/.02a744470e2ca072.stage | /tmp/huntress-autoit/.1dda60946801d965.stage |
Note the ATT&CK answers: the descriptor's own attack= field wins over the language default. AutoIt/AutoHotKey as an interpreter is T1059.010, and that is what the adapter carries for .au3, but these fixtures are graded on their terminal intent - remote transfer gives T1105, decode-and-dispatch gives the generic T1059, and the benign parser gives the literal none. Follow the descriptor, not the file extension.
Why it was already solved
autosolve_script.py line 44 already mapped the AutoIt layer before the release:
(('.ahk', '.au3'), 'autoit', 'T1059.010', (), dict(_C, concat=r'\.|&')), # AutoHotKey & AutoIT
AutoIt/AutoHotKey was on the lane's prepared next-layer candidate list, the descriptor spine (Material_/Codec_/load/decode/recover/inspect/plan_ + HUNTRESS_ANALYSIS_V1) was unchanged, and all four carrier encodings were known. So the new language cost nothing: the adapter fingerprinted the extension, pulled the descriptor, and answered all 855 fields for Single, Time Trial and Mastery with no extension needed. This is the fifth consecutive day the "self-describing descriptor" hypothesis has held.
Pairing theme
The "matched set / which pairing is genuine" framing is the Day 9 stage-relationship graph carried forward, now expressed across the AutoIt pair: next_language plus stage_profile decide whether a hand-off is real (cross_language_handoff -> same_language/cross_language graph values) or merely "going through the motions". The three carried-over PowerShell scenario fields (resource-scenario, collection-scenario, impact-scenario -> resource_hijacking / synthetic_collection / synthetic_impact) and the Day 11 browser credential-store question (collection-attack -> the credential-store sub-technique, not the broader Data from Local System) were unchanged from yesterday.
Takeaways
- Prepared candidate layers pay off: AutoIt was on the lane's candidate list and already in
LANGS, so a brand-new language produced a 3-second three-mode sweep. HB64(hex-of-base64) andFRAG(dot-fragmented hex) are the two carrier spellings worth keeping: both appear only as the header tag, and both reduce to one line of Python.- The descriptor's
attack=value always overrides the interpreter's own technique ID. T1059.010 is the right answer for "classify the AutoIt interpreter itself", never for "what does this specimen do".