mobile: 31 days of a puzzle inside an app

the mobile lane asked one thing all month: an ANS- code decoded from the puzzle json an apk carries in its assets. the manifest, deep links, intents and each day's story were flavor; the codec named in the puzzle was the challenge. it cycled through seven names for two weeks, picked up names outside the cycle from day 15, moved the timed tiers to a code the app's own dex computes from day 23, and ended on framed, keyed and layered codecs. the solver became a codec registry with keys fitted from the ANS- prefix, a bounded layer peel and a dalvik interpreter that runs the app's code.

  • 31challengesone a day, days 1 to 31
  • 93flagssingle, time trial and mastery
  • 2.6 smedian singletime trial 5.1 s, mastery 8.5 s; release to flag
  • 77%flags by autopilot72 of 93
  • 227inventory items46 graded, 181 not yet graded
  • 21 of 27covered at releasedays 5 to 31; 5 partial, 1 cold start
mobile: release to the single's flag, per day
0:002:305:007:3010:00day 1: 2h48m, unattributed (off the scale)1day 2: 9:56, lane fixerday 3: 0:30, helper3day 4: 0:49, helperday 5: 1:16, helper5day 6: 0:02, autopilotday 7: 0:04, autopilot7day 8: 0:04, autopilotday 9: 0:04, autopilot9day 10: 0:04, autopilotday 11: 0:01, autopilot11day 12: 0:02, autopilotday 13: 0:02, autopilot13day 14: 0:02, autopilotday 15: 0:02, autopilot15day 16: 0:03, autopilotday 17: 0:19, lane fixer17day 18: 0:01, autopilotday 19: 0:02, autopilot19day 20: 0:02, autopilotday 21: 0:01, autopilot21day 22: 0:02, autopilotday 23: 0:02, autopilot23day 24: 0:02, autopilotday 25: 0:02, autopilot25day 26: 0:29, parentday 27: 0:02, autopilot27day 28: 14:12, parent (off the scale)day 29: 0:17, lane fixer29day 30: 0:03, autopilotday 31: 0:03, autopilot31
  • autopilot
  • lane fixer
  • helper
  • parent
  • other

the scale stops at ten minutes; a dashed bar is a day beyond it (hover for the time). a bar's colour is the role on the flag: autopilot claims each single at release and runs the lane solver, a lane fixer edits that solver, a helper works a single nobody owns, a parent routes and submits only a value it already holds. day 1 opened at 11:28, so its bar counts from 09:00 and includes the late start.

how the family evolved

one answer id, answer: ANS- and 16 uppercase hex digits, decoded from assets/huntress_puzzle.json ({schema, codec, data[, key]}). the question never changed shape.

days 1 to 7, the cycle. plain (exported trust), reverse (deep link drift), base64 (intent injection), a keyed xor (pending intent relay), split parts put back by an order list (backup spill), hex (provider paths) and a caesar shift on the answer text (webview bridge).

days 8 to 14, the cycle again, out of order. reverse, base64, split, xor, hex, base64 and caesar under stories of cleartext traffic, trust managers, network config, local stores, sqlite, a keystore and crypto (cleartext session, trust everyone, network config split, local secrets, sqlite assumptions, keystore misuse, predictable crypto). every story was only a story; autopilot took all seven singles at release.

days 15 to 22, names outside the cycle. native twice (native library secrets, jni boundary), indexed, one character per index (name collapse), a keyed xor again (encrypted strings), reflection, a name read by its answer's shape (reflection maze), dynamic, the code in a second dex (loaded later), then reverse (debug me if you can) and base64 (root of trust). from day 15 mastery mixed older codecs into its sets.

days 23 to 26, a computed code. the timed tiers moved to a second puzzle schema with no data: the app's dex computes the code, so a dalvik interpreter went in on day 23 (hook line and signal). the single kept a first-schema codec on those days, and day 26 shipped the app as a base apk and a feature split (split decision).

days 27 to 31, frames, tables and layers. base64 on the single with an eleven-set mastery of older codecs (dependency surface); a framed packet with a checksum trailer, 14:12 to the single (protocol reconstruction); a byte table keyed by its index (white box key recovery); reversed bytes under a walking xor key and base64 (layered protections); on day 31 the computed code reached the single too (mobile arena gauntlet).

the codec in each tier's puzzle json, read from the saved apks (a mastery cell counts its sets and their distinct codecs):

day the single's codec the time trial's mastery post
1 plain plain 3 sets, all plain exported trust
2 reverse reverse 3 sets, all reverse deep link drift
3 base64 base64 3 sets, all base64 intent injection
4 xor xor 3 sets, all xor pending intent relay
5 split split 3 sets, all split backup spill
6 hex hex 3 sets, all hex provider paths
7 caesar caesar 3 sets, all caesar webview bridge
8 reverse reverse 5 sets, all reverse cleartext session
9 base64 base64 5 sets, all base64 trust everyone
10 split split 3 sets, all split network config split
11 xor xor 3 sets, all xor local secrets
12 hex hex 3 sets, all hex sqlite assumptions
13 base64 base64 3 sets, all base64 keystore misuse
14 caesar caesar 3 sets, all caesar predictable crypto
15 native native 5 sets, 3 codecs native library secrets
16 native native 5 sets, 5 codecs jni boundary
17 indexed indexed 5 sets, 5 codecs name collapse
18 xor xor 5 sets, 5 codecs encrypted strings
19 reflection reflection 5 sets, 5 codecs reflection maze
20 dynamic dynamic 5 sets, 5 codecs loaded later
21 reverse reverse 5 sets, 4 codecs debug me if you can
22 base64 base64 6 sets, 4 codecs root of trust
23 xor analysis 7 sets, all analysis hook line and signal
24 split analysis 8 sets, all analysis emulator alibi
25 hex analysis 9 sets, all analysis integrity signals
26 split analysis 10 sets, all analysis split decision
27 base64 base64 11 sets, 8 codecs dependency surface
28 packet analysis 12 sets, all analysis protocol reconstruction
29 table analysis 13 sets, all analysis white box key recovery
30 layered analysis 14 sets, all analysis layered protections
31 analysis analysis 15 sets, all analysis mobile arena gauntlet

coverage

this lane's row of the coverage grid. autopilot answered 21 of the 27 singles from day 5 on with no gaps row; the table lists the rest. closed in runs from autopilot's gaps row to the single's flag.

mobile: the single at release, by day
12345678910111213141516171819202122232425262728293031mobileday 1: no release pass, single at 2h48m (unattributed). no release pass yet: autopilot's first release pass was day 5day 2: no release pass, single at 9:56 (lane fixer). no release pass yet: autopilot's first release pass was day 5day 3: no release pass, single at 0:30 (helper). no release pass yet: autopilot's first release pass was day 5day 4: no release pass, single at 0:49 (helper). no release pass yet: autopilot's first release pass was day 5 (on day 4 it first ran at 11:44:40, after the singles)day 5: partial, single at 1:16 (helper). 1 field (answer): no decoding path foundday 6: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 7: covered, single at 0:04 (autopilot). nothing: autopilot answered every field at releaseday 8: covered, single at 0:04 (autopilot). nothing: autopilot answered every field at releaseday 9: covered, single at 0:04 (autopilot). nothing: autopilot answered every field at releaseday 10: covered, single at 0:04 (autopilot). nothing: autopilot answered every field at releaseday 11: covered, single at 0:01 (autopilot). nothing: autopilot answered every field at releaseday 12: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 13: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 14: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 15: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 16: covered, single at 0:03 (autopilot). nothing: autopilot answered every field at releaseday 17: partial, single at 0:19 (lane fixer). 1 field (answer): no decoding path foundday 18: covered, single at 0:01 (autopilot). nothing: autopilot answered every field at releaseday 19: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 20: partial, single at 0:02 (autopilot). 1 field (answer): no decoding path foundday 21: covered, single at 0:01 (autopilot). nothing: autopilot answered every field at releaseday 22: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 23: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 24: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 25: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 26: partial, single at 0:29 (parent). 1 field (answer): the split parts did not joinday 27: covered, single at 0:02 (autopilot). nothing: autopilot answered every field at releaseday 28: cold start, single at 14:12 (parent). 1 field (answer): no decoding path foundday 29: partial, single at 0:17 (lane fixer). 1 field (answer): no decoding path foundday 30: covered, single at 0:03 (autopilot). nothing: autopilot answered every field at releaseday 31: covered, single at 0:03 (autopilot). nothing: autopilot answered every field at release
  • covered at release
  • partial, under 3 min
  • cold start, 3 to 15 min
  • long, over 15 min
  • no release pass or unknown

hover a cell for the time, the role on the flag and what was missing. grey on days 1 to 4: the release pass started on day 5.

day at release what was missing closed in
5 partial 1 field (answer): no decoding path found 1:15, helper
17 partial 1 field (answer): no decoding path found 0:16, lane fixer
20 partial 1 field (answer): no decoding path found 0:00, autopilot
26 partial 1 field (answer): the split parts did not join 0:26, parent
28 cold start 1 field (answer): no decoding path found 14:09, parent
29 partial 1 field (answer): no decoding path found 0:14, lane fixer

the tooling that grew for it

the lane's tool page, android apk puzzle extraction, opens with what the solver did by day 3 and adds a section for every day its code changed:

every post, in order

times run from the 09:00 et release to each tier's flag; day 1 opened at 11:28 and counts from 09:00.

day post single time trial mastery
1 exported trust 2h48m 2h48m 2h50m
2 deep link drift 9:56 10:37 10:40
3 intent injection 0:30 0:50 1:00
4 pending intent relay 0:49 1:22 1:31
5 backup spill 1:16 1:23 1:25
6 provider paths 0:02 0:04 0:06
7 webview bridge 0:04 0:08 0:11
8 cleartext session 0:04 5:22 5:23
9 trust everyone 0:04 0:06 0:08
10 network config split 0:04 0:06 0:07
11 local secrets 0:01 0:03 0:05
12 sqlite assumptions 0:02 0:03 0:05
13 keystore misuse 0:02 0:04 0:06
14 predictable crypto 0:02 0:05 0:07
15 native library secrets 0:02 0:04 0:06
16 jni boundary 0:03 0:05 0:08
17 name collapse 0:19 0:34 0:36
18 encrypted strings 0:01 0:04 0:05
19 reflection maze 0:02 0:04 0:06
20 loaded later 0:02 0:04 1:08
21 debug me if you can 0:01 0:03 0:06
22 root of trust 0:02 0:06 0:08
23 hook line and signal 0:02 5:22 5:24
24 emulator alibi 0:02 0:04 0:07
25 integrity signals 0:02 0:05 0:07
26 split decision 0:29 3:04 3:18
27 dependency surface 0:02 0:04 0:08
28 protocol reconstruction 14:12 16:15 16:18
29 white box key recovery 0:17 1:12 1:16
30 layered protections 0:03 0:05 0:09
31 mobile arena gauntlet 0:03 0:05 0:09

{generated by a clanker}