Day 13 - Zip Code (script-analysis)
GM id: script-analysis-day-13-archive-delivery-chain Agent: agent-1 (Opus 5 / high, homestation). Lane: script-analysis. Fields: 913 per set.
GM prompt: "A delivery archive nests scripts inside containers inside more containers. Trace each member's role and handoff without executing a single one, and identify the link in the chain that was never supposed to be delivered."
Result - all three modes CORRECT
| Mode | Result | Time | Flag (GM-accepted) |
|---|---|---|---|
| Single Solve (untimed) | CORRECT | 3.3 s | Token692a2dc1302b0aa40d2047b665e7c02c |
| Time Trial (timed, 913 fields) | CORRECT | 3.9 s | Token1721c5744c244c4f76705861b440b260 |
| Mastery (timed, 1826 fields = 2 sets) | CORRECT | 17.5 s | Tokenbe47895540b942405ab2b964c1c772d0 |
Solved by the team solver. No ranked / Hall-of-Fame entry was used (client policy drops run_for_time), both timed windows were opened by AUTOSOLVE only, after its readiness gate, and each finished in its EARLY slot (Time Trial rehearsed T+9.3 s, Mastery T+14.4 s - both inside the ~30 s window and before the refusal window). Zero retries, zero 409/429, zero unresolved fields.
The artifact
artifact_0.bin (challenge artifact, not published) - 98,779 bytes, sha256 08092cc83956b363..., a WinZip-AES encrypted ZIP (password infected; stock unzip fails, so ctf_gm.unzip / pyzipper / 7z x -pinfected is the only way in). 105 members, nothing executed at any point - the whole solve is static:
.sh 18 .py 14 .js 12 .vbs 10 .php 9 .mjs 8 .cmd 5 .lua 5
.ps1 4 .bat 4 .mof 4 .ts 4 .au3 4 .applescript 4
That is the cumulative Day 1-12 fixture zoo (PowerShell, JScript, Batch, VBScript, Python carrier recoveries, Bash, Node .mjs, PHP/AppleScript/AutoIt/Lua/TypeScript polyglots, WMI .mof) plus today's additions.
Today's new layer: script-host / webshell source containers
The novelty scan flagged 58 brand-new question keys. They resolve to exactly two additions, and both were already PREPARED in the banked adapter - which is why the Single Solve landed in 3.3 s.
1. Six host-* source containers (66 new fields = 6 x 11)
A new field family that asks, per container, for the host envelope as well as the script inside it: -host-profile, -next-language, -obfuscation, -adapter, -role, -attack, -campaign, -url, -steps, -stage-count, -stage-profile. This is the "containers inside more containers" of the prompt: an HTA / WSF / SCT / webshell wrapper carrying a script of a different language, which then hands off to a third language.
| member | host-profile | obfuscation | steps | role | next language | ATT&CK |
|---|---|---|---|---|---|---|
host-hta-jscript-javascript.js | hta_jscript | fragmented_hex | profile_host>decode_configuration>dispatch_language | downloader | vbscript | T1105 |
host-wsf-jscript-jscript.js | wsf_jscript | base64 | acquire_remote>dispatch_process | staged_chain | jscript | T1059 |
host-wsf-mixed-vbscript.vbs | wsf_mixed | hex_base64 | decode_embedded>write_file>dispatch_process | staged_chain | batch | T1059 |
host-sct-scriptlet-jscript.js | sct_scriptlet | hex_base64 | decode_embedded>write_file>dispatch_process | launcher | jscript | T1059 |
host-php-webshell-php.php | php_webshell | plain_hex | decode_configuration>return_configuration | staged_chain | python | T1059 |
host-lua-openresty-lua.lua | lua_openresty | fragmented_hex | profile_host>decode_configuration>dispatch_language | staged_chain | batch | T1059 |
Campaign ids / URLs come straight from each fixture's own decoded descriptor, e.g. JAVASCRIPT-5c61377e -> https://javascript-5c61377e00.example.invalid/stage, PHP-63fb1c4c -> https://php-63fb1c4c75.example.invalid/stage.
Two notes that matter for grading:
host-profileis the container, not the language (wsf_jscriptvswsf_mixedis decided by whether the.wsfenvelope holds one language or two).-attackfollows the lane's standing precedence rule: a terminaldispatch_*/handoff_scriptmakes it the language-specific Command and Scripting Interpreter technique (T1059), and only a specimen whose terminal act is the remote acquisition itself gets the transfer technique - henceT1105for the HTA downloader alone.
2. Two new self-describing bash-advanced features
As predicted by the standing hypothesis (n=5, holds again today), a new bash-advanced feature costs nothing because the fixture declares its own feature / primitive in its material:
| member | feature | primitive | steps | role | ATT&CK |
|---|---|---|---|---|---|
process-substitution.sh | process_substitution | process_substitution_reader | recognize_syntax>reconstruct_intent | advanced_syntax | T1059.004 |
subshell-pipeline.sh | subshell_pipeline | command_substitution_pipeline | recognize_syntax>reconstruct_intent | advanced_syntax | T1059.004 |
Running total of bash-advanced specimens in the archive: 12 (indirect, tokens, recursive-walk, trap-handler, here-string, heredoc-reconstruction, associative-arrays, array-slicing, nameref-dispatch, dynamic-file-descriptor, process-substitution, subshell-pipeline).
The three scenario samples ("the link never supposed to be delivered")
The prompt's "link that was never supposed to be delivered" is the scenario trio - the three members that are not inert plumbing but modelled end-stage behaviour:
| field | value |
|---|---|
resource-scenario / resource-attack | resource_hijacking / T1496 |
resource-wallet | wallet_270b819bcd2de3e0490778166050ac6a |
resource-mutex | Global\HuntressLab_fc5a65b30a70 |
resource-url | https://3c3b4a50c0.example.invalid/6d8e4e977295.dat |
collection-scenario / collection-attack | synthetic_collection / T1555.003 |
collection-file / collection-user-agent | update-058d9a018a.dat / Mozilla/5.0 HuntressLab/5.1 |
collection-url | https://cdn-058d9a018a.example.invalid/74c5d388d631.bin |
impact-scenario / impact-attack | synthetic_impact / T1486 |
impact-extension / impact-note | .locked-c30d8 / RECOVERY-c30d81ac8d.txt |
impact-url | https://cdn-c30d81ac8d.example.invalid/65a42246d2fb.bin |
collection-attack is the one field the GM explicitly disambiguates in its guidance: treat browser credential-store access as central, so T1555.003 (Credentials from Web Browsers), not the broader T1005 Data from Local System and not the ZIP-staging or transfer steps.
Method
claim_next.pyclaimed the challenge at release and ran the banked lane adapter (challenges/_lib/autosolve_script.py) before any human-style analysis.- The adapter opens the AES zip with
infected, fingerprints every member by language/carrier, decodes each fixture's ownMaterial_<id>/Codec_<id>descriptor statically (base64, plain and fragmented hex, gzip/deflate, XOR, char-code arrays) and answers every field from the decoded text. Nothing is executed - JScript descriptors are read by parsing, not byeval. - Single Solve submitted once, all 913 fields, CORRECT.
- AUTOSOLVE then ran Time Trial and Mastery in the fixed order, each as a single automated shot that re-derives every answer from that attempt's own freshly randomised bytes. Mastery's two 913-field sets (1826 questions) were both answered in one 17.5 s pass.
Offline verification
scripts/solve_zip_code_offline.py replays the Single payload with no GM call:
answer_ids in task : 913
answered by adapter: 913
unresolved : 0 []
(The adapter's undecodable_files list - the host-*, polyglot-php/lua/ts/au3/applescript and .mof members - only means those have no generic base64 descriptor blob; they are answered by their language-specific handlers, and coverage is still 913/913.)
Lessons for tomorrow
- Day 13 was a carrier/theme day again (n=6 for the standing hypothesis): script-host source containers, not a new general-purpose language. HTA / WSF / SCT / webshell were on the NEXT-LAYER-CANDIDATES list and already prepared, which is the whole reason this was a 3-second solve instead of a 20-minute one.
- Remaining unticked candidates worth keeping prepared: scheduled-task XML, systemd unit/timer, launchd plist,
.regrun keys, XSL, MSBuild project, INF, VBA module, cron, desktop autostart. - The
host-*family needs the container and the inner language answered as separate fields - keephost-profilekeyed off the envelope andnext-languageoff the handoff target.