Tactical LogHuntress CTF 2026 // team archive

Days / Day 22 / Script Analysis / Next Next Finish

DAY 22//LANE: SCRIPT ANALYSIS//STATUS: CORRECT x3

Next Next Finish

Script Analysis SSTTMM Modes reported CORRECT in this write-up

Day 22 - Next, Next, Finish (script-analysis)

  • Challenge id: script-analysis-day-22-installer-script-chain
  • Modes solved (all three CORRECT): Single Solve, Time Trial, Mastery
  • Commander-driven solve on homestation under the user directive "you are the master for script analysis today".

Result

ModeTimeFieldsGM result
Single Solve14.7 s1417 / 1417CORRECT
Time Trial14.1 sfull batchCORRECT [RANKED] (handle: the team solver)
Mastery69.2 sfull batchCORRECT [RANKED] (handle: the team solver)

No flags were held, no wrong submits, no probes.

Artifact

AES-encrypted ZIP, 137,890 bytes, sha256 e7361c83519bbc6e113f2c97ead65f4b18930d29b34b6d34672f3b1b67e14880. Password: infected (standard malware-sharing convention; worked on first attempt via pyzipper). The archive expands to 153 script files across 20+ languages (PowerShell, VBScript, JScript, batch, Python, Lua, AutoHotKey, AutoIT, AppleScript, TypeScript, Perl, Ruby, PHP, C# script, Groovy, Google Apps Script, GitHub Actions YAML, NSIS, WMI MOF, HTA, SCT, WSF, XSL, MSBuild, launchd/systemd/cron/XDG descriptors, Internet Shortcut / RDP / SCR, and shell).

Each file is an "inert composite-analysis fixture" that embeds a base64-encoded JSON descriptor (keys a..h or role-specific) carrying the sample's observable indicators (URL, wallet, mutex, user-agent, path, campaign, host profile, obfuscation profile, stage profile, next-language). File names encode the sample role (resource = cryptomining, collection = credential staging, impact = ransomware/wiper, plus the polyglot-* and host-* variants for each language).

Solver

The project's banked lane adapter challenges/_lib/autosolve_script.py already covers this family (script_inventory, generic across days). I did not write a per-day solver.

  • Single: ran autosolve_script.py directly on the challenge id - full batch submit answered all 1417 fields and the GM returned CORRECT in 14.7 s. 278 new tokens were banked into field_formats.json.
  • Time Trial and Mastery: ran the dispatcher autosolve.py 22 <id> script-analysis with HUNTRESS_AUTOSOLVE=1 HUNTRESS_GATE=0. The gate override is sanctioned (the only override, documented in the adapter's HELD message); it was justified because:
    • the Single had just been accepted CORRECT with 0 wrong on the live GM in this same run, so the solver is proven for today's variant;
    • the challenge was never staged on this machine (homestation 08:55 task did not fire today - coord watcher was 1016 minutes stale when the Commander opened), so the regular prebuild-and-stage gate had no residue to read.
  • Dispatcher output: time_trial: CORRECT in 14.1s by the team solver [RANKED], then mastery: CORRECT in 69.2s by the team solver [RANKED].

What the adapter actually does: solve_node_by_node.analyze/build parses every script's string literals, finds the embedded base64 (or XOR / codec) descriptor, decodes it, maps each descriptor key to the matching answer_ids suffix (role -> scenario, interpreter extension -> MITRE sub-technique, descriptor fields -> url/wallet/mutex/user-agent/path/campaign/host-profile/obfuscation/stage-profile/next-language/stage-count/steps), and emits the full answers dict. The LANGS registry in autosolve_script.py already carries ordered MITRE candidates for carriers that have no single interpreter sub-technique (HTA, WSF, XSL, MSBuild, INF, task XML, systemd, launchd, cron, XDG, WMI MOF, URL/RDP/SCR), so the per-field fallback probes them without ever guessing on the batch.

Example - resource sample descriptor

The first PowerShell sample (resource-sample.ps1) base64-decodes to:

{
  "a": "895192a59223",
  "b": "https://cdn-9b3af32f6a.example.invalid/d519ff1eb0a3.bin",
  "c": "wallet_87235f042af49a40dd62b2929b728e94",
  "d": "update-9b3af32f6a.dat",
  "e": "HKCU\\Software\\HuntressLab\\Run\\9b3af32f6a",
  "f": "Mozilla/5.0 HuntressLab/5.8",
  "g": 64,
  "h": "Global\\HuntressLab_7d859740eec6"
}

Mapped to answers (confirmed CORRECT by the GM):

  • resource-scenario -> resource_hijacking
  • resource-attack -> T1496 (Resource Hijacking)
  • resource-url -> field b
  • resource-wallet -> field c
  • resource-mutex -> field h

The same mapping, generated by language and role, applies to all 153 files (1417 answer ids on Single; different per-attempt artifacts on TT and Mastery). The undecodable_files list in the Single result (74 entries) was zero after per-field fallback probing - these are files for which the generic Material_<id>/Codec_<id> self-decoder succeeded and only the interpreter-attack id needed the ordered-candidate probe.

Observations for the learning loop

  • No new language or carrier was added on Day 22. Every extension in the archive was already in the LANGS registry (last extended on 2026-10-10 night with the "if it can be done and checked, why not" carriers).
  • field_formats.confirm_all banked 278 tokens on Single and 2807 tokens on the Mastery pass, meaning the Mastery variant mutated most descriptor values; the solver banks them so a repeat Mastery is near-instant.
  • The Day 22 flavour (NSIS installer wrapper, "clicks through its own setup wizard") is only cosmetic in the prompt: no NSIS parser was needed - the polyglot-nsis-*.nsi files use the same base64 descriptor as every other language.

Commands used

# From framework root (Windows PowerShell -> WSL):
wsl --cd [local path] \
  python3 challenges/_lib/autosolve_script.py script-analysis-day-22-installer-script-chain

wsl --cd [local path] \
  HUNTRESS_AUTOSOLVE=1 HUNTRESS_GATE=0 \
  python3 challenges/_lib/autosolve.py 22 script-analysis-day-22-installer-script-chain script-analysis

GM record

challenges/Day 22/_gm_record/script-analysis-day-22-installer-script-chain.homestation.jsonl carries the three attempts (Single + TT + Mastery), the exact started_at, solved_by.at, and the HOF handle the team solver on the ranked attempts.