chunk norris: png dimensions from the ihdr chunk
huntress ctf 2026, day 5 · metadata forensics · all three modes
tools used: archive and metadata toolkit and timed-attempt pipeline.
the evidence zip held png images, and the new field family asked for each image's dimensions as declared in its IHDR chunk. the prompt talked about walking chunks, verifying declared geometry and finding an image with an unexpected structural passenger, but the graded field was only the declared size.
the key insight: IHDR is always the first chunk at a fixed offset, so the answer is two big-endian 32-bit integers read straight from the header, then serialized exactly as WIDTHxHEIGHT.
challenge shape
answer ids follow the lane's <member stem>-<property> pattern, here d05-png-NNN-png-ihdr-dimensions-pixels, asked as "determine png.ihdr_dimensions_pixels for panel-nnn.png." with the same generic exact_text syntax as day 4. the single and the time trial carried one image; mastery carried one nested zip per set, each with its own image.
artifact anatomy
a png starts with an eight-byte signature, then a sequence of chunks, each length (4) | type (4) | data (length) | crc32 (4) with the crc taken over type and data:
89 50 4e 47 0d 0a 1a 0a signature
00 00 00 0d 49 48 44 52 length 13, "IHDR"
ww ww ww ww hh hh hh hh width, height (big-endian u32)
08 02 00 00 00 bit depth, color type, compression,
filter, interlace
.. .. .. .. crc32
every graded image was a clean IHDR IDAT IEND file: no ancillary chunks, no bytes after IEND, valid crcs.
extraction method
def image_size(data: bytes) -> tuple[int, int]:
if data.startswith(b"\x89PNG\r\n\x1a\n") and data[12:16] == b"IHDR":
w, h = struct.unpack_from(">II", data, 16)
return w, h
... # gif, bmp, webp and jpeg headers follow
from a shell, xxd -s 16 -l 8 panel-NNN.png shows the same eight bytes (width, then height), and file panel-NNN.png prints them as decimal.
the property name was new, but the toolkit already routed any unregistered *-dimensions-pixels field to this header reader (prepared on day 4), so the existing solver answered it without a code change.
validation and answer formatting
the reader checks the signature and that the first chunk really is IHDR before trusting offset 16; anything else falls through to the other image formats and finally raises. the answer is width first, a lowercase ascii x, then height, with no spaces, the same serialization as day 4's jpeg frame size. the declared size is what was asked, not the decoded pixel grid.
failed approaches and pivots
none. on day 5 we also took the prompt's hints seriously, since a cumulative lane tends to ask about what earlier prompts described. the toolkit now walks the whole chunk list and checks each crc. it also reports chunk types outside the png specification (the likely "passenger"), counts bytes after IEND, and checks the declared geometry: the inflated IDAT stream must be height * (1 + row bytes) long for a non-interlaced image.
automation across modes
the same three-step pipeline as every metadata day: split the field into member stem and property, resolve the member in the (possibly nested) archive, compute the property from the raw bytes. nothing was day-specific.
lessons
- an unregistered property should resolve by meaning (
*-dimensions-pixels) rather than crash; that fallback is why this day needed no code. - read prompts for what they describe, not only for what they ask; the chunk walk was prepared from a hint, before any field asked for it.
solve times
public scoreboard · day 5 finish: 1st
- 0:02.3singlefirst blood, +14.7 s
- 0:53.4time trial#2 of 240 solvers
- 0:55.2mastery#2 of 238 solvers
- #1team day finishof 188 teams that cleared it, in 3:35.1
- #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 | 1s | 1 | 0 |
| time trial | 51s | 1 | 0 |
| mastery | 2s | 1 | 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 | 0.8s | 5.2s |
| mastery | 6.0s | 0.9s | 5.1s |
where we got stuck
- nothing analytical. the time trial waited about 50 seconds before its attempt started, a scheduling delay on our side; the attempt itself took 0.8 seconds.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}