2026-10-03
Day 3: Table Manners
Lane: Binary reverse engineering. Modes: Single Solve, Time Trial, Mastery. Result: all CORRECT.
The challenge
- ID:
binary-reverse-engineering-day-03-managed-configuration - Prompt: "A .NET binary is a database wearing a trench coat. Read the CLI metadata tables — methods, strings, signatures — and find where the configuration is stashed.
ilspycmdandstringsare your friends."
What we were given
- One
evidence.zip(SHA-2569d7a0aee73c18b68c60bb8b8785ec1a6407ca54f9ae7c2df0d0890adcc37d777). AES encrypted, passwordinfected. - From today on, each file name starts with its day number (
dNN--…, heredNN--variant-NNN). The bundle grows by one day each day. - Members, in ZIP central-directory order:
d01--variant-000.elf(15,616 bytes): a plain C program, like Day 1.d02--variant-000.elf(16,640 bytes): the C++ program from Day 2.d03--variant-000.dll(9,216 bytes) andd03--variant-000.runtimeconfig.json(98 bytes). New today: a .NET (managed) assembly.
What was asked
- 24 questions in Single Solve and Time Trial. Mastery had two sets, so 48.
- Each question says exactly how to read the answer from the bytes, and the exact answer format.
- Question types:
- Hashes: SHA-256 of each member.
- d01 header facts: ELF class, byte order, object type, CPU architecture.
- d01 structure: section names in order, symbol names, and the disassembly of
recover_code. - d02 config: format, XOR key, CRC32, and the
campaign,domain,urlandmutexfields. - d03 (new): the combined format (container + runtime), the PE/COFF Machine, the CLR metadata version string, and the
campaign,domainandmutexconfig fields.
How it was solved, start to finish
- Started the attempt (
bdf56aea-7f01-4d7c-b02e-7ff3b9e5236c) and downloaded the zip. - Extracted it with 7z and hashed each member over its exact bytes.
- d01: read the ELF header and tables straight from the bytes:
- ELF64, little endian,
EXEC, x86-64. - Section names come from the section header string table. Symbol names come from
.symtab. Both were sent as compact JSON arrays. recover_codeis at0x401126and is 6 bytes long:mov eax,0x1fa4thenret. The answer was["mov eax,0x1fa4","ret"].
- d02: the
.hcfgsection holds the only CRC32-validHCFGrecord (the Day 2 layout: magic, version, XOR key, payload length, CRC32, payload). XOR key0x31, CRC320x6149aa00, campaignBR02-000. - d03 (new): the .NET DLL also hides an
HCFGrecord, as plain bytes in the file. The same decode works.hcfg.pyalso reads the CLR version (atBSJB+ 16) and the PE Machine field:
=== HCFG valid records in d03--variant-000.dll ===
offset=0x1898 version=1 xor_key=0x41 plen=367 crc32=0xcb6cf23f
JSON: {"campaign":"BR03-000","domain":"node-000.day03.example.invalid","ipv4":"198.51.100.60","url":"https://node-000.day03.example.invalid/api/BR03-000","mutex":"Global\\HuntressBR03_000", ...}
=== CLR metadata version (BSJB+16, len=12) ===
version: 'v4.0.30319'
=== PE/COFF Machine ===
machine: 0x014c (332)
So d03 is PE32/.NET, machine i386, metadata version v4.0.30319.
- Submitted all 24 answers (
answers.json). Result:CORRECT. - Time Trial: a fresh variant (zip SHA-256
62c2a27adbc9e69cc3fba445ac2eac8558e2e7eb068a23b61f1c6993924301ca). Here d03 wasvariant-001: XOR key0x42, CRC320x55fbb381, campaignBR03-001. Same steps. Result:CORRECT. - Mastery: two deliveries, 48 answers (
set1.*,set2.*).
- The fresh timed variants were used up (
unique_attempts_remaining: 0). So the bundle was finished in untimedpracticemode, which the rules allow and which still gives the flag. - Zips:
evidence-01.zip(SHA-256dee83e7c6271273f7ab5f7383864365d62d948faa35ce7654a87197adbcd9558) andevidence-02.zip(SHA-2563e7a0f66b583ea7153a5ee1344d38b2c2409011ff353f90cb518075974418128). - In set 2, d02 was a Windows PE (
.exe, PE32+, machine0x8664). Format and machine are read per file, so the same steps worked. - Result:
CORRECT.
Tools from this day, in ~/.config/huntress/solvers/:
hcfg.pyprints every CRC32-validHCFGrecord, the CLR version and the PE machine (the output above).build_tm_answers.pybuilds exactly these 24 answers for thevariant-000file names.solve_tm.pyis the general version ("General Table Manners solver"). Later days kept extending it.
Results
| Mode | Result | Flag |
|---|---|---|
| Single Solve | CORRECT | 6cc99f5d631b392122f08923c47db36d |
| Time Trial | CORRECT | 3719598ae32e2983bd5ece6364e63c41 |
| Mastery | CORRECT (practice) | d26a68dfa228dff3300a99d24bc3b537 |
Each mode folder has the zip, the extracted members, the answers, commands-output.md, and the raw Game Master responses. No screenshots were taken.
Reproduce
Offline, from this folder (no Game Master needed):
mkdir -p /tmp/d03/deliveries && cp single-solve/evidence.zip /tmp/d03/deliveries/delivery-1.zip
python3 -I ~/.config/huntress/solvers/solve_tm.py single-solve/start-response.json /tmp/d03/deliveries /tmp/d03/answers.json
Checked on 2026-11-01: this gives the same 24 answers as answers.json. It only parses the files (7z, Python, objdump) and never runs them.
Live driver (start → download → solve → submit):
python3 -I ~/.config/huntress/solvers/run_tm.py binary-reverse-engineering-day-03-managed-configuration single_solve /tmp/d03
This needs a Game Master token in ~/.config/huntress/gm-token.
Day 3: BOM Voyage
Lane: Metadata forensics. Modes: Single Solve, Time Trial, Mastery. Result: all CORRECT.
The challenge
- ID:
metadata-forensics-day-03-bom-voyage - Prompt: "A text export pipeline produced files that look identical in an editor but behave differently in scripts. Find the encoding landmines before they break the investigation."
What we were given
- Single Solve:
(SHA-256metadata-day-03-evidence.zip207c74a175c574a31325c983afa860b59477238a9d1d9e075de7cfabadf0a285). It holds.evidence-note.txt - Time Trial:
(SHA-256metadata-day-03-evidence.zip941acab50c421c8f13be3e9cecde35b70398cb721fba007f991daf51ef7752bb). It holds.roster.csv - Mastery:
(SHA-256delivery.ziped058314fa5b627b19a14df5e1f7b9f44f6cd440569df4ede1466ccd8ef011bb). It holds two inner zips: metadata-day-03-evidence-01.zipwith.evidence-note-copy.txtmetadata-day-03-evidence-02.zipwith.export-001-with-bom.txt
What was asked
- One field type:
text.utf8_bom_present. Does the file start with a UTF-8 byte order mark? - The answer must be
trueorfalse. - Answer IDs:
- Single Solve:
evidence-note-text-utf8-bom-present. - Time Trial:
roster-text-utf8-bom-present. - Mastery:
set1.evidence-note-copy-text-utf8-bom-presentandset2.d03-text-001-text-utf8-bom-present.
How it was solved, start to finish
The method: a UTF-8 BOM is the three bytes EF BB BF at the very start of the file. Look at the first bytes, nothing else.
- Started the attempt.
- Single Solve: attempt
c1635b66-ed73-46e4-95ca-8b16fdbf7613. - Time Trial: attempt
387ded04-0b44-4cc8-bf13-ff738b335905, deadline2026-10-20T14:07:46.302Z. - Mastery: attempt
6e9b2b4c-ccf9-4b2f-8bce-a00503b819e5, deadline2026-10-20T14:08:24.894Z.
- Downloaded the zip for the mode.
- Dumped the first line of hex of each assigned file:
unzip -p metadata-day-03-evidence.zip evidence-note.txt | xxd -g 1 | head -n 1
- Read the result:
| File | First bytes | BOM? |
|---|---|---|
evidence-note.txt | 43 61 73 (Cas) | false |
roster.csv | 68 6f 73 (hos) | false |
evidence-note-copy.txt | 43 61 73 (Cas) | false |
export-001-with-bom.txt | ef bb bf | true |
Mastery output from commands-output.md:
00000000: 43 61 73 65 20 45 31 38 39 43 46 30 34 46 41 34 Case E189CF04FA4
00000000: ef bb bf 45 78 70 6f 72 74 20 31 0d 0a 43 61 73 ...Export 1..Cas
- Submitted the answers. Mastery sent
falsefor set 1 andtruefor set 2 (seeanswers.json). - The Game Master replied
CORRECTand returned the flag.
Results
| Mode | Result | Flag |
|---|---|---|
| Single Solve | CORRECT | 696cb5b99ae3d0f7d682bfd2faf6fe9a |
| Time Trial | CORRECT | 35a871fa3af9775bc418dd77d1963e7e |
| Mastery | CORRECT | 9aad61f691e39c200044abf96246fc45 |
Each mode folder has start-response.json, answers.json, submit-response.json, final-status-response.json, commands-output.md and the evidence files. No screenshots were captured.
Reproduce
python3 -c 'import sys; print(open(sys.argv[1],"rb").read(3) == b"\xef\xbb\xbf")' mastery/export-001-with-bom.txt
run_meta.py later uses the same rule: -utf8-bom-present returns 'true' if fb[:3]==b'\xef\xbb\xbf' else 'false'.
Day 3: Intent Injection
Lane: Mobile (Android). Modes: Single Solve, Time Trial, Mastery. Result: all CORRECT.
The challenge
- ID:
mobile-android-day-03-intent-injection - Prompt: "Trace attacker-controlled extras through an internal navigation boundary."
What we were given
- One Android APK per case, 12,704 bytes each. Inside:
AndroidManifest.xml,classes.dex,assets/huntress_puzzle.json, andMETA-INF/signing files. - Single Solve:
intent_injection.apk, caseMOB-03-8006859B52, packagecom.huntress.mobile.d03.v8006859b5257, SHA-2563127e3a93000b594fe4555747aed236f6f77bd8daaf2be41cfe60849bf8708aa. - Time Trial: saved as
time-trial/intent_injection.apk, caseMOB-03-495E12E539, SHA-2569d8ba27fa0f621e0eefd583b222cfb50338972b4a4698514233023888315b8bf. - Mastery:
intent_injection_mastery_01.apkto_03.apk(casesMOB-03-DEB2E5C362,MOB-03-66ED139540,MOB-03-3F36C0F510). Their puzzles are saved aspuzzle-01.jsontopuzzle-03.json.
What was asked
- One question per case: return the exact backend answer code for that case.
- Answer format:
ANS-plus exactly 16 uppercase hex characters. - Answer IDs:
answerfor Single Solve and Time Trial.set1.answertoset3.answerfor Mastery. - Time limits: Time Trial 50 seconds (deadline
2026-10-21T13:39:20.177Z), Mastery 110 seconds.
How it was solved, start to finish
The app was never installed or run.
- Started the attempt (Single Solve attempt
8f06326b-c40e-406c-9154-df2592096e6a) and downloaded the APK. - Read the DEX.
ChallengeActivity/IntentInjectionread the nested-Intent extrasanswerandcase_idand relay them to a local evidence endpoint. The exact answer code for the case ships in the puzzle asset. - Read the puzzle asset. New codec today:
base64.
$ unzip -p intent_injection.apk assets/huntress_puzzle.json
{"schema":"huntress.mobile.puzzle.v1","day":3,"codec":"base64","data":"QU5TLUU0QzhFMzVBOEE5QjVENjc="}
$ python3 -c "import json,base64;print(base64.b64decode(json.load(open(\"huntress_puzzle.json\"))[\"data\"]).decode())"
ANS-E4C8E35A8A9B5D67
- Submitted
ANS-E4C8E35A8A9B5D67underanswer. The Game Master repliedCORRECT. - Time Trial (attempt
49099f34-9d29-496e-9343-aa3c2472dd55): same steps.QU5TLTMxQjJEQUI4QTE3QTRENkY=→ANS-31B2DAB8A17A4D6F. - Mastery: decoded each APK's asset, then sent all three in one submit:
set1→ANS-D013DAB129ECC3D5set2→ANS-BD7AF4BB5C717C53set3→ANS-CA2248A783994E07
Results
| Mode | Result | Flag |
|---|---|---|
| Single Solve | CORRECT | 28408637c686113abeb550f2f35b2e2b |
| Time Trial | CORRECT | 9aee2378c9ea2f81087eaafbdbff31a9 |
| Mastery | CORRECT | 313d84460563cdb9d54ae5c5d5298c75 |
Each mode folder has the APK(s), the puzzle JSON, answers.json, start-response.json, submit-response.json, final-status-response.json and a commands-output.md. No screenshots were taken.
Reproduce
Offline, from the saved puzzle:
cd single-solve
python3 -I -c 'import json,base64;print(base64.b64decode(json.load(open("huntress_puzzle.json"))["data"]).decode())'
A fresh run with the later lane driver (it handles the base64 codec):
python3 -I ~/.config/huntress/solvers/run_mode.py mobile-android-day-03-intent-injection single_solve mobile_puzzle /tmp/d03
This needs a Game Master token in ~/.config/huntress/gm-token. It only unzips the APK and reads the asset. It never installs or runs the app.
Day 3: Expansion Pack
Lane: Script analysis. Modes: Single Solve, Time Trial, Mastery. Result: all CORRECT.
The challenge
- ID:
script-analysis-day-03-batch-expansion - Prompt: "cmd.exe expands percents, delays expansion, CALLs twice, and bends FOR loops into indirection. Reconstruct every variable exactly as the interpreter would see it — without executing a single line — then follow the staged chain to its final handoff."
What we were given
- One
evidence.zip(21,255 bytes, SHA-256ce1a3fb0a5425cfa0794fc7b7f09ec93b86b74544ad93283a55ebe0ee3fcc23e). Password:infected. It is saved assingle-solve/artifact.zip. - 21 files, one matched set per language:
- 3 PowerShell samples (same as Days 1–2).
- 9 JScript files (
*.js) and 9 Batch files (*.bat/*.cmd). - Both use the same nine file names as Day 2 (
host-context= benign,memory-stage= stager, ...relay-handoff= relay). - New today: Batch. Each Batch file has:
- a
rem HUNTRESS-CODEC:line and arem HUNTRESS-MATERIAL:blob, and - an
:HUNTRESS_INSPECTblock that rebuilds the config fromset "fragmentN=..."lines.
What was asked
- 159 questions: 15 PowerShell, 72 JScript (as Day 2), 72 Batch (8 per file).
- Batch questions per file:
- Role and ATT&CK ID, from the decoded behavior graph.
- Adapter, campaign ID, the graph topology (
single,same_language,cross_language,diamond,fan_out), and one role-specific indicator. - Expansion: how the config line is rebuilt:
percent,delayed,call_doubleorfor_indirect. - Control: what makes that line run:
direct,goto_dispatch,call_subroutine,for_once,if_guardorerrorlevel_guard.
How it was solved, start to finish
- Single Solve came first (attempt
09742ae4-ec42-45eb-abb9-be3726163501, solved 2026-10-20 14:11:52 UTC).
- The saved
submit-response.jsonis an earlier try that came backincorrect. - The final status shows all 159 answers accepted.
- Its
answers.jsonanddecoded-descriptors.jsonbecame the ground truth for the general solver below.
- PowerShell: same Base64 JSON decode as Day 1. Scenario and ATT&CK are fixed per file (
T1496,T1555.003,T1486). - JScript: each file hides its config with a different scheme (constant XOR, cumulative XOR, reverse then XOR, Base64 with key bytes, delimiter splits).
- Reversing each one by hand would break on the next variant.
- So
jsdec.jsloads the fixture into a Nodevmsandbox that only has pure-computation globals (norequire,process, file system orWScript). - It then calls the fixture's zero-argument functions until one returns the
{steps, adapter, values, stageCount}object.
- Batch: the config sits in plain text in the INSPECT fragments. Real lines from
memory-stage.bat:
:HUNTRESS_INSPECT
SeT "fragment0=v=1|steps=acquire_remote,dispatch_memory|adapter"
SeT "fragment1==certutil.exe|stage_count=1|graph=single|next_la"
SeT "fragment2=nguage=none|campaign_id=campaign-0054cccedb|url="
SeT "fragment3=https://0054cccedb.example.invalid/2c9a4553.dat"
batchdec.py joins fragment0..N in order, strips the ^ escape carets, and splits on | and =. Simplified:
frags = {int(n): v for n, v in re.findall(r'set\s+"fragment(\d+)=([^"]*)"', text, re.I)}
cfg = ''.join(frags[i] for i in range(len(frags))).replace('^', '')
fields = dict(p.split('=', 1) for p in cfg.split('|'))
- If the fragments are missing, it decodes
HUNTRESS-MATERIALinstead. It tries each candidate transform and keeps the one whose text starts withv=1|steps=.
- Source-style questions come from regexes on the file text, not from the config:
- Expansion:
for ... do call set→for_indirect;call set "x=%%a%%"→call_double;!a!!b!→delayed;%a%%b%→percent. - Control, checked in this order:
if errorlevel→errorlevel_guard;if defined→if_guard;call :labelanywhere →call_subroutine;gototo a non-standard label →goto_dispatch; elsedirect.
- Role and ATT&CK:
- Role follows the last step of the graph (
write_recovery_note= impact,upload_remote= uploader, and so on). - ATT&CK follows the role: benign →
none, stager and downloader →T1105, dropper →T1140, uploader →T1041, impact →T1486. Launcher, chain and relay take the language's Command and Scripting Interpreter ID:T1059.003for Batch,T1059.007for JScript. - The solver reproduced all 159 Single Solve answers. Two fixes were needed first:
- the collection "first file" must skip URLs;
- function style is read from the hex-named main functions (
decode_,recover_,inspect_), not from helpers.
- Time Trial (50-second deadline), the hard part:
- The first try ran out of time, and one Batch file used a transform the decoder did not know (
base64_literal|add_7). Fix: make the INSPECT fragments the main Batch source. - The next full submit came back
incorrect. Key finding: a partial submit is graded on its own. It returnsaccepted(plus aremainingcount) only if every answer in it is right, and accepted answers stay on the attempt. diag_submit.pysent the answers grouped by language and field. Only three groups failed:batch:control,jscript:function-styleandjscript:invocation-style.bf.pythen tried each allowed value for those fields, one answer at a time, until each wasaccepted. That finished the attempt (CORRECT) and showed the right values.- The learned values fixed the three detectors:
.apply(meansapply;var X = function NAME(meansnamed_expression;call :LABELanywhere on a line beatsgoto_dispatch.
- A later fresh attempt (the one saved in
time-trial/) was solved in one shot. Seetime-trial/commands-output.md. - Mastery (80-second deadline): 318 questions in two sets. The download is a plain outer zip holding two AES zips,
evidence-01.zip(set 1) andevidence-02.zip(set 2).solve_mastery.pyextracts both, runssolve_expansion.pyon each and adds thesetN.prefix. All 318 answers were then sent in one submit (gm.sh submit). It wasCORRECTon the first try. Seemastery/commands-output.md.
Results
| Mode | Result | Flag |
|---|---|---|
| Single Solve | CORRECT | 901513368dd3bb9ec59f5b2ffe37e67b |
| Time Trial | CORRECT | 3fb2c5c8b2b61427b1588c458a07cbd7 |
| Mastery | CORRECT | 0cc3ef5343f2c1d33bc3b45bcd60d4b8 |
Time Trial: attempt 4d3e8f10-39be-4369-bcca-478aab3c36c1, archive SHA-256 7d3e7825af480dccc34e6392f37304094cbb6c38cdc2df30e02c6af40002d063. Mastery: archive SHA-256 91e813196c048e46b017e8edde6bee0139c2450ad5ab2467c296fc74125e1829. Each mode folder keeps its archive, the extracted files and the Game Master JSON. No screenshots were taken.
Reproduce
Offline, from the repo root:
mkdir -p /tmp/d03/ex
7z x -pinfected -o/tmp/d03/ex repos/day-03-script-analysis-batch-expansion/single-solve/artifact.zip
python3 -I tools/solve_expansion.py /tmp/d03/ex \
repos/day-03-script-analysis-batch-expansion/single-solve/start-response.json /tmp/d03/answers.json
This rebuilds all 159 saved answers (re-checked on 2026-11-01). The tools/ copy decodes JScript with jsdec_py_emu.py (pure Python) instead of the Node sandbox, so nothing from the samples runs. The scripts used on the day (solve_expansion.py, diag_submit.py, bf.py, solve_mastery.py) are in ~/.config/huntress/solvers/. The submit helpers need a Game Master token.
Day 3: Hex Appeal
Lane: Warmups. Modes: Single Solve, Time Trial, Mastery. Result: all CORRECT.
The challenge
- ID:
warmups-day-03 - Prompt: "Two characters, sixteen values. This one's straightforward if you think in nibbles."
- New encoding: hexadecimal
What we were given
- One blob file per mode (Single Solve and Time Trial), or a zip of multiple blobs (Mastery).
- The blob is a single encoded string.
- The decoded text ends with a slug like
word-word-word-NNto submit.
How it was solved
- Decode the blob. Hex-decode the blob (pairs of hex chars to ASCII bytes).
- The decoded text contains a message ending in the answer slug.
- Submit the slug to the Game Master.
Results
- Single Solve:
CORRECT. slug:nimble-scooter-boulder-88. Flag:fe409374e47ae3c7e2a738e7f49e513b. - Time Trial:
CORRECT. slug:jovial-island-copper-93. Flag:9f59006d6f12a49b5a2eba5727ef3d2a. - Mastery:
CORRECT. 13 slugs. Flag:7ea9222d856fc9fd3db9a74b0317ebbf.