Tactical LogHuntress CTF 2026 // team archive

Days / Day 30 / Binary RE / Campaign Trail

DAY 30//LANE: BINARY RE//STATUS: CORRECT x3

Campaign Trail

Binary RE SSTTMM Modes reported CORRECT in this write-up

Day 30 - Campaign Trail (binary-reverse-engineering)

All three modes CORRECT.

  • single_solve flag: Token1b280bfe17571044b2a167d4357aa10e
  • time_trial flag: Token627a0e29763c885c37c564ad37ff6cb3
  • mastery flag: Tokenbbdb3a10de9716eb6c542c634a8221b2

What the day added

Day 30 is a cumulative recap (d01..d30, 14 fields/day = 420 TT ids, x14 sets = 5880 MM ids). The only genuinely new layer is d30 itself: a NEKO-bytecode campaign of three role-named members per set:

  • d30--sample-loader.n
  • d30--sample-library.n
  • d30--sample-config.n

Each file is a Neko bytecode module (magic NEKO) carrying an appended HGRF trailer that uses the same 12-byte layout as HCFG, differing only in the magic bytes:

+0  magic 'HGRF'   (4)
+4  version        (1)
+5  xor key        (1)
+6  length le16    (2)
+8  crc32  le32    (4)
+12 xor-encoded utf-8 json payload (length bytes)

The JSON is a huntress.binary.campaign_graph.v1 capsule with graph.nodes, graph.edges, graph.load_order, graph.config.{campaign,domain,ipv4,url,mutex, windows_path,linux_path,service,host_model} and a per-member this_role tag (loader / library / config). All three role copies carry an identical graph.config when this_role is removed, as the question text asserts.

The 7 unresolved fields on first run

The lane spine already answered 413/420 of the Single's ids by catalog (d01..d29 shapes + d30 structural / artifact_sha256_N / Neko-based elf-less inventory). The 7 residues were the new d30 class:

  • d30_analysis_loader_format -> neko_bytecode
  • d30_analysis_library_format -> neko_bytecode
  • d30_analysis_config_capsule_format -> neko_bytecode
  • d30_analysis_config_campaign -> BR30-000 (from loader.n HGRF)
  • d30_analysis_config_domain -> ...example.invalid (from loader.n HGRF)
  • d30_analysis_config_mutex -> Global\HuntressBR30-000(from loader.n HGRF)
  • d30_analysis_config_url -> https://... (from loader.n HGRF)

...and in MM each set also asks the capsule_graph_config_* variants that select the config role member instead of the loader. All variants decode to the same shared config object; what differs is only which role-named member supplies the HGRF record.

Patch (in binary_wip.py answer(), before the member-picker)

Added an explicit d30 fast path that:

  1. Picks the role-named member by basename (endswith(role + '.n')).
  2. Classifies it as neko_bytecode when its first 4 bytes are NEKO.
  3. For a analysis_config_capsule_graph_config_<name> suffix, reads the HGRF record from the config member and returns graph.config.<name>.
  4. For a plain analysis_config_<name> suffix (non-capsule), reads the HGRF record from the loader member and returns graph.config.<name>.

The HGRF scan starts from rfind(b'HGRF') (the question says "scan backward from the end"), bounds-checks the length, CRC32-validates the XOR-decoded payload before accepting it, and never guesses.

With that one extractor added, SS went 420 sure / 0 probe / 0 unknown; the runner then opened the TT and MM windows and both passed CORRECT first try.

Lessons + D+1

  • Specific extractor/opcode missing: HGRF capsule records on NEKO bytecode. The 12-byte layout is identical to HCFG except the magic. Keep this in the shared kit so a future day that reuses the exact capsule can be answered by the spine without a clone edit.
  • Stall time: ~5 minutes. The 413 pre-answered fields proved the lane spine was already correct for d01..d29 end to end; the only build was the d30 role picker + HGRF decode.
  • Keep in clone vs port to shared: port. The role-named-member picker (loader.n / library.n / config.n) and the HGRF decoder are both generic shapes - a future day that bundles role-named members or ships a capsule with a 1-byte-magic twin of HCFG will reuse exactly this logic. The clone copy is fine for today; shared autosolve_binary should pick up a small _hgrf_scan sibling of _hcfg_scan and wire the three format-suffixes + the two config-suffix families.

Reproducibility

  • Artifact: single_0.bin (challenge artifact, not published) (evidence.zip, zip-crypto password infected).
  • Members (per set): d30--sample-{loader,library,config}.n, each ~20.9 KB.
  • HGRF trailer offsets observed on single_0.bin: loader off=20173 ver=1 key=0xF1 crc=0xC41517FB, library off=20182 ver=1 key=0xF1 crc=0x8E1F39C8, config off=20209 ver=1 key=0xF1 crc=0x76C956E7.
  • All three decode to the same graph.config modulo the this_role field.