bom voyage: a byte-order mark is a byte predicate
huntress ctf 2026, day 3 · metadata forensics · all three modes
tools used: archive and metadata toolkit and timed-attempt pipeline.
the new field asked whether a named text member had a utf-8 byte-order mark. it sounds like encoding detection; the exact property is simpler.
the key insight: utf-8 bom presence means the raw file starts with the three bytes EF BB BF. the grader asks about bytes, so bytes are the evidence.
challenge shape
one field per member, <stem>-text-utf8-bom-present, and nothing else: the single and the time trial asked one field each, the mastery one per set across two sets. the question read Determine text.utf8_bom_present for <member>., an enum with accepted_values true and false. in mastery the field's stem need not equal the archive member's name.
extraction method
import codecs
answer = "true" if body.startswith(codecs.BOM_UTF8) else "false"
do not decode and re-encode first: many text apis consume or normalize the bom, and a string starting with U+FEFF does not prove which byte encoding was delivered.
member resolution uses layered evidence:
- the question's explicit artifact path when it names a real member;
- an exact parsed stem;
- the stem's numeric identifier as a whole token;
- the only extracted member when the assignment is otherwise unambiguous.
validation and answer formatting
lowercase true or false. the resolver ignores container zips and compares basenames deliberately: fuzzy substring matching could confuse 001 with 1001 or pick an archive instead of its member.
failed approaches and pivots
a field such as d03-text-001-text-utf8-bom-present contains -text- in both its stem and its property. splitting at the first occurrence chose the wrong member; matching the longest known property suffix (equivalently, splitting from the right) fixed it.
automation across modes
the day-3 solver is only a dispatch loop: split the field into stem and property, resolve the raw member, call the property registry. the same code still answers day-1 hashes and day-2 magic prefixes in every mode.
lessons
- measure byte properties on bytes, before any decoding.
- parse field grammars from the right when stems can contain property words.
solve times
public scoreboard · day 3 finish: 1st
- 0:28.6singlefirst blood, +26.3 s
- 0:38.4time trialfirst blood, +42.8 s
- 1:03.0masteryfirst blood, +1:42
- #1team day finishof 199 teams that cleared it, in 14:26.6
- #2single 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 when we started working on the mode to its flag.
| mode | work to flag | attempts | wrong submits |
|---|---|---|---|
| single | 22s | 1 | 0 |
| time trial | 5s | 1 | 0 |
| mastery | 24s | 3 | 0 |
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 | 6.0s | 0.3s | 5.7s |
| mastery | 5.0s | 1.0s | 4.0s |
where we got stuck
- member stems could themselves contain
-text-, so splitting the field at the first occurrence picked the wrong property; split from the right. - in mastery the question stem did not always match the member name, so resolution fell back to the question's artifact path and the stem's number.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}