deflated ego: a zlib line, and a slug that ran past its shape

huntress ctf 2026, day 22 · warmups · all three modes

tools used: decoding search engine, replay and analysis and timed-attempt pipeline.

"another familiar wrapper. this one needs a little inflation." day 21 hid the message in base64 of a gzip member; day 22 hid it in base64 of a zlib stream. the single and the time trial are one such blob each; the mastery is a zip of 108 blobs of twenty-two numbered lines, one line per day of the event, whose newest line is the same base64-of-zlib.

the key insight: a reading that a checksum verifies must never lose to a reading that merely fits the answer's expected shape. one mastery blob's message ran past the slug shape the question describes, and a search that wanted that shape picked a garbage reading of the last line to get it.

challenge shape

single         1 field    the slug one blob's decoded text instructs
time trial     1 field    the same, a fresh blob
mastery      108 fields   one slug per blob, 108 blobs of 22 numbered lines

the field is message (setK.message in mastery).

artifact anatomy

the single's blob is base64; decoded, it starts with 78 9c (a zlib header: deflate, a 32k window, check bits that make the first two bytes a multiple of 31) and ends with the adler-32 of the decompressed text. in the mastery each blob is NN<tab>chunk lines: one message of about 110 characters split evenly over 22 lines, line k encoded by day k's encoding. line 21 is base64 of gzip, line 22 base64 of zlib.

extraction method

the decoding engine reads line k by day k's row first and searches only a line past the schedule. the zlib line needed no new transform: b64inflate, added for day 21's gzip line, does a strict base64 decode and then tries the decompressors in turn, zlib first, then gzip, reading the first clean result as text. in stdlib terms:

raw = base64.b64decode("".join(chunk.split()), validate=True)
text = zlib.decompress(raw).decode("utf-8")   # zlib header; wbits=31 for gzip

validation and answer formatting

the question states the slug's shape ("three lowercase words separated by hyphens, followed by a hyphen and a two-digit number; submit the slug only"), and the search used that shape to tell a right reading of the newest line from a wrong one. that worked for 107 of 108 mastery blobs. in one, the decoded message ran on past the slug for a few more characters, all of it from a zlib stream whose adler-32 checked. the shape check (nothing may follow the slug but a word boundary) refused that reading, and the next search tier read line 22 as reversed base64, which left a shaped slug followed by = and garbage. that shaped slug was rejected in every form.

the fix: a slug followed by more -token groups counts as found, so the verified reading wins; the answer goes out as two candidates, the whole token first and its shaped prefix second, both tried in the same attempt. a timed attempt later on day 22 held two such blobs, and the grader took the whole token both times.

core = shape.pattern.removesuffix(r"(?![A-Za-z0-9-])")
extended = re.compile(core + r"(?:-[a-z0-9]+)+(?![A-Za-z0-9-])")

failed approaches and pivots

the first fix was a handler that appended the whole token to the candidates for that one challenge; the tier was flagged on the next attempt, whose fresh bytes had no such blob, so the handler's candidate was never graded. the general rule replaced it the same day, and the later timed attempt confirmed it. the pattern showed up only in timed attempts, 3 of the 432 timed mastery blobs we saw on day 22, and in none of the 3,706 untimed blobs we pulled for testing that day: testing on untimed variants alone would not have found it.

automation across modes

the same engine reads the single and time-trial blobs and every mastery blob; in the mastery only a line past the schedule is searched, and now a message that runs past the expected shape is answered with both readings.

lessons

solve times

public scoreboard · day 22 finish: below top 10

  • 0:03.4single#3 of 177 solvers
  • 0:06.6time trial#3 of 170 solvers
  • 3:59.6mastery#42 of 166 solvers
  • #26team day finishof 145 teams that cleared it, in 10:39.5
  • #2single orderof the day's 7, 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 when we started working on the mode to its flag.

mode work to flag attempts wrong submits
single 1.5s 1 0
time trial 1.1s 1 0
mastery 231.9s 2 18

the clock

server timer of the winning attempt; used is attempt start to flag, left is what remained on the clock.

mode time available time used time left
time trial 30.0s 1.1s 28.9s
mastery 125.0s 0.9s 124.1s

where we got stuck

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

{generated by a clanker}