dependency surface: a stored code behind a reachability story
huntress ctf 2026, day 27 · mobile android · all three modes
tools used: android apk reader and timed-attempt pipeline.
"prove which generated dependency flaw is reachable from an exported mobile surface." day 26's split bundle gave way to one small apk again, twelve kilobytes, with the usual puzzle description inside it and one answer code to recover.
the key insight is that the prompt's story and the graded field are two different things: the field is the same ANS- code the lane has asked since day 1, stored in the apk under a named codec, and the puzzle reader that handles every codec the family has used answered it without reading a line of the app's code.
challenge shape
single 1 field the ANS- code, sixteen uppercase hex digits
time trial 1 field the same on a fresh apk
mastery 11 fields one code per set: ten apks and one split bundle
artifact anatomy
an apk is a zip: AndroidManifest.xml (binary xml), classes.dex, the resources and, in this family, a json puzzle description under assets/ that names how the code is encoded. the day's json was schema v1 with codec base64 and the code's base64 inline in data (28 characters):
{"schema": "huntress.mobile.puzzle.v1", "day": 27, "codec": "base64",
"data": "<28 base64 characters>"}
the manifest's exported, browsable activity (a custom-scheme deep link) was the story's "surface" and played no part in the answer. the mastery's sets carried earlier days' codecs: reverse (three times), xor, hex twice, caesar, native, dynamic, day 26's split bundle, and today's base64.
extraction method
the apk reader parses the puzzle json, maps the codec name to its steps through the alias table (base64 to b64) and runs them on data; by hand:
import base64, json, zipfile
puzzle = json.loads(zipfile.ZipFile("dependency_surface.apk")
.read("assets/huntress_puzzle.json"))
code = base64.b64decode(puzzle["data"]).decode()
base64 has been in the alias table since day 3.
validation and answer formatting
the code must match ANS- followed by sixteen uppercase hexadecimal characters; the reader checks the pattern before submitting and treats anything else as a crash to fix. ANS-802A91C1FCCB2F7E for the single's apk, ANS-2F5233A432900F0D for the time trial's and ANS-7AA32F2FD188C460, ANS-66793ECA13FD97B2, ANS-9A3C573081CA4EBD and eight more for the mastery's eleven sets.
failed approaches and pivots
none on day 27.
automation across modes
the same reader answered every mode; the mastery's eleven sets were solved as their bytes arrived, 7 milliseconds of solve in total. almost all of the three windows was the server's: provisioning and serving the sets.
lessons
- read the graded field, not the prompt's narrative: the story about exported surfaces had no field behind it.
- a reader that handles every codec the family has used wins the day a known codec returns with a new story.
solve times
public scoreboard · day 27 finish: below top 10
- 0:03.5single#2 of 150 solvers
- 0:05.3time trial#2 of 148 solvers
- 0:09.2mastery#2 of 145 solvers
- #38team day finishof 127 teams that cleared it, in 34:43.1
- #2single orderof the day's 8, 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 | 0.9s | 1 | 0 |
| time trial | 0.7s | 1 | 0 |
| mastery | 1.1s | 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 | 50.0s | 0.7s | 49.3s |
| mastery | 350.0s | 1.1s | 348.9s |
where we got stuck
- nowhere on day 27: the puzzle reader read the code on the first run of every mode.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}