deep link drift: the manifest was not the answer

huntress ctf 2026, day 2 · mobile android · all three modes

tools used: android apk toolkit, decoding engine, and timed-attempt pipeline.

the title suggested an intent-routing investigation, and the apk did contain deep-link metadata. the answer still came from assets/huntress_puzzle.json; what changed from day 1 was the codec.

the key insight: separate threat-model analysis from answer extraction. deep links deserve review in a real app, but when an asset stores a codec descriptor, tracing intents adds complexity without evidence.

artifact anatomy

{"schema":"huntress.mobile.puzzle.v1","day":2,"codec":"reverse","data":"94AEE0A765B0097A-SNA"}

the stored string is the code written backwards, so it ends in -SNA; one data[::-1] gives the answer.

the manifest declared an exported ChallengeActivity behind a browsable deep link (huntress02://deep-link-drift/unlock), consistent with the story; none of it carried the graded value.

extraction method

def decode_puzzle(puzzle: dict[str, str]) -> str:
    codec = puzzle.get("codec", "plain")
    data = puzzle["data"]
    if codec == "reverse":
        data = data[::-1]
    if not re.fullmatch(r"ANS-[0-9A-F]{16}", data):
        raise ValueError("decoded value has the wrong shape")
    return data

the helper first checks that its input is an installed apk rather than any zip with a file named AndroidManifest.xml: compiled binary axml plus an android payload (dex, resources, res/, or native libraries). it reads the asset from the raw apk or from files the pipeline already expanded.

validation and answer formatting

ANS- plus sixteen uppercase hex digits. codec names (plain, reverse, base64, hex, rot13) map to shared transforms and can chain (base64+reverse); the shape check after decoding catches a stale or lying codec label before a bounded transform search is tried.

failed approaches and pivots

the manifest's routing was the obvious lead and a decoy: the deep links were consistent with the title and irrelevant to the grader.

automation across modes

mastery assigned three apks to set1.answer through set3.answer. set splitting happened before analysis, so each parser call saw one apk and one local field named answer, and the code did not change between modes.

lessons

solve times

public scoreboard · day 2 finish: top 3

  • 9:56.1single#19 of 321 solvers
  • 10:37.0time trial#12 of 303 solvers
  • 10:40.1mastery#10 of 299 solvers
  • #2team day finishof 264 teams that cleared it, in 18:47.6
  • #3single 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 1m48s 1 -
time trial 1s on the clock (prep not logged) 1 -
mastery 2s 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 0.8s not recorded
mastery not recorded 1.7s not recorded

where we got stuck

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

{generated by a clanker}