power play: structured indicators hidden in powershell

huntress ctf 2026, day 1 · script analysis · all three modes

tools used: script-analysis toolkit and timed-attempt pipeline.

three powershell samples represented resource hijacking, collection and impact. each hid its configuration in an inline base64 literal that decoded to json, so the task was to recover and classify a descriptor, not to run powershell.

the key insight: classify indicators by the shape of their values, not by json key positions, and let the lane decide the behavioral labels.

challenge shape

fifteen fields, five per sample: a scenario label, an att&ck technique, and three indicators that differ per lane (urls, wallet, mutex, user agent, staging file, extension, recovery note).

artifact anatomy

the zip used winzip aes (AES-256 Deflate in 7z l -slt) with the conventional malware-analysis password infected, so 7z x -pinfected evidence.zip is the extraction step. it held one <lane>-sample.ps1 per lane (resource, collection, impact). each script wrapped one function around a call like:

$encodedMaterial = [Convert]::FromBase64String('<base64>')
...
$descriptor = $decodedText | ConvertFrom-Json

and a $nativeTypes array of .net type literals that hints at the behavior: [Diagnostics.Process], [Microsoft.Win32.Registry], [Net.WebClient] in the resource sample; [IO.File], [IO.Compression.ZipFile], [Net.HttpWebRequest] in the collection sample; [IO.File], [Security.Cryptography.Aes], [IO.DirectoryInfo] in the impact sample. the decoded descriptor's keys are single letters (a, b, c, ...), and which letter holds which indicator differs per sample.

extraction method

m = re.search(r"FromBase64String\('([A-Za-z0-9+/=]+)'\)", source)
descriptor = json.loads(base64.b64decode(m.group(1)))

keys changed across variants, so the solver flattens scalar and list values and classifies by shape: http strings are urls, wallet_... a wallet, Global\... a mutex, Mozilla/... a user agent, a leading-dot token an extension, RECOVERY-*.txt a ransom note, and the first ordinary .dat or .zip name the collection staging file. the question asks for the "first exact filename indicator", and the list that holds it also carries the browser store names Login Data and Cookies, which are not filenames with an extension, so the first name.dat / name.zip match is the answer. the mutex is escaped as Global\\... inside the json text; json.loads turns it into one backslash, and the one-backslash form is what the grader accepted.

validation and answer formatting

behavior supplies the semantic labels:

lane scenario att&ck characteristic evidence
resource resource_hijacking T1496 wallet, mutex, url
collection synthetic_collection T1555.003 browser data, staging file, user agent
impact synthetic_impact T1486 encrypted extension, recovery note, url

the scenario fields are an enum of exactly those three values; the technique fields take T#### or T####.### (the question's syntax line). indicator values are exact_text, submitted verbatim with no quotes; $nativeTypes is a cross-check, not an answer source.

failed approaches and pivots

macos unzip cannot open winzip-aes archives; extraction moved to 7-zip. the collection technique was the only wrong field at first: generic data collection (T1005) sounded defensible, but the sample specifically read browser Login Data and cookies, so the grader expected credentials from web browsers (T1555.003). probing fields one at a time isolated that mismatch without discarding the fourteen correct values.

automation across modes

the time trial used a fresh descriptor with the same patterns, and mastery doubled the answer set (two zips, thirty fields named set1.<field> and set2.<field>); the same shape classifier handled both because nothing depended on key order or specimen-specific values. no sample ever runs, which removes host risk and environmental variability.

lessons

solve times

public scoreboard · day 1 finish: below top 10

  • 2h45msingle#77 of 492 solvers
  • 2h55mtime trial#71 of 437 solvers
  • 2h59mmastery#70 of 428 solvers
  • #54team day finishof 358 teams that cleared it, in 3h04m
  • #4single orderof the day's 5, by our flag time

from the public scoreboard: 09:00 et release to our flag; the table below counts from our attempt's start.

from our first start of the mode to its flag; earlier preparation was not recorded.

mode work to flag attempts wrong submits
single 16m59s 1 -
time trial 1s on the clock (prep not logged) 1 -
mastery 1s on the clock (prep not logged) 1 -

the clock

server timer of the winning attempt; used is attempt start to flag, left is what remained on the clock. day 1-2 deadlines were not logged per attempt.

mode time available time used time left
time trial not recorded 1.4s not recorded
mastery not recorded 1.1s not recorded

where we got stuck

flags: single, time trial, and mastery [redacted].

{generated by a clanker}