Tactical LogHuntress CTF 2026 // team archive

Days / Day 29 / Script Analysis / Make It Make Sense

DAY 29//LANE: SCRIPT ANALYSIS//STATUS: CORRECT x3

Make It Make Sense

Script Analysis SSTTMM Modes reported CORRECT in this write-up

Make It Make Sense (Day 29, script-analysis)

Challenge id: script-analysis-day-29-makefile-variant-diff Lane: script-analysis Agent: agent-1 (the team solver) Date solved: 2026-10-29

Flags (all three modes CORRECT)

ModeFlagTimingRanked
Single SolveTokene56d3ebde626f7c4e5e084bc9f1d599f18.7 sno
Time TrialToken2e8dda3a4fe1656df47be52d7238cdf935.1 syes
MasteryToken218111bd0cc658306e716a52ff1a7f7d271.0 syes

Prompt

"Unseen Makefile variants and their prior-language build chains differ in layered variables and target flow. Diff them carefully, and keep only the tradecraft conclusions that hold up across every variant."

Artifacts

  • artifacts/single_0.bin (challenge artifact, not published) 159,137 bytes, SHA256 0ba75808186d63e2269cd7d69f2d6d2835cd6425300ce2aa9a389204218a1525
    • AES-encrypted ZIP, password infected, 16+ member script fixtures (ps1, js, vbs, cmd, mjs, sh, csx, php, pl, rb, lua, gs, groovy, jsp, applescript, ahk, au3, dockerfile, exs, escript, gd, github_actions, groovy, jsp, lua, mof, msbuild_xml, nsis, perl, php, ruby, terraform_hcl, typescript, wmi_mof, plus polyglot-makefile-*.mk).
  • artifacts/time_trial_0.bin (challenge artifact, not published) 178,178 bytes - same shape, AES-encrypted ZIP (pw infected), 16 members.
  • artifacts/mastery_0.bin (challenge artifact, not published) 2,338,988 bytes - plain ZIP wrapping 13 nested evidence-*.zip members (bundle-of-variants form).

Recognize - deterministic path

  1. file single_0.bin -> Zip archive data ... method=AES Encrypted. Standard script-lane carrier.
  2. peel_any.py <artifact> -> zip(aes, password infected) with a per-member type table (powershell, javascript, vbscript, batch, bash, node, polyglot-*). That listing alone identifies the family.
  3. Family fingerprint: 50-cell polyglot-script library with per-field Material_<id>/Codec_<id> descriptors - the script_inventory family handled every day since D8. Today's new carrier is polyglot-makefile-*.mk (benign-utility, downloader, launcher, staged-chain) added into the same spine; descriptor keys unchanged.
  4. The in-tree autosolve_script adapter already covers all of that; the only load-bearing step at runtime is the AES unlock with the hard-coded CTF password infected.

Solve - what the adapter actually did

  • claim_next.py agent-1 -> single_0.bin pre-fetched; script_wip.py cloned as the per-CID solver copy.
  • autosolve_script.run() iterated the ZIP members, peeled the carrier, ran solve_fields_ext on the decoded text, and submitted a single batch:
    • single_solve: CORRECT in 18.7s.
    • field_formats.confirm_all banked 44 new token(s) -> fresh makefile-variant field tokens learned on SS and reused for TT/MM.
  • Time Trial (178 kB ZIP, 16 polyglot members): field_formats.confirm_all banked 1833 new token(s) -> CORRECT 35.1s [RANKED].
  • Mastery (2.34 MB bundle of 13 nested evidence-*.zip): field_formats.confirm_all banked 23829 new token(s) -> CORRECT 271.0s [RANKED].

No per-field extractor edits were needed; the SS adapter answered TT and MM unmodified.

Hiccups (and how to avoid them next time)

Timeline (actual runs for this CID):

  • 09:00 SS CORRECT (18.7s)
  • 09:00 TT HELD - "time_trial contract has no answer IDs in the release assignment" (per-release contract not staged yet; the lane-specific script-analysis-day-29-makefile-variant-diff/time_trial had no answer_ids until a GM.start was made)
  • 09:13 Rerun: TT HELD - "solver copy changed after the prediction was recorded" (mode_prediction.json was bound at 09:00 before any later solver-copy touch invalidated the gate hash)
  • 09:17 Rerun: TT HELD - "gate unavailable - prior mode CORRECT does NOT authorize time_trial without a verified current-mode contract; D29 P0-A fail-closed"
  • 09:28 Rerun: TT HELD - "current GM status unavailable (TimeoutError)" (transient GM read timeout on readiness check)
  • 09:31 Rerun: TT CORRECT (35.1s, RANKED), MM CORRECT (271.0s, RANKED)

Root causes:

  1. The release assignment arrived with empty answer_ids for the timed modes (n_questions: 0), so the pre-start prediction journal couldn't be populated and the gate kept failing closed. The adapter has a documented bypass ("prediction gate skipped - letting GM.start() provide answer IDs; using the SS adapter"), but it requires the prediction gate to pass once first.
  2. The solver-copy hash churned because the capability layer rewrote script_wip.py on each run start; mode_prediction.py bind had captured the pre-restart hash. The gate correctly refuses to open the timed window on a hash mismatch (prevents spending a variant on an unverified solver).
  3. A single transient TimeoutError on GM.status burned one more restart cycle - the gate is deliberately conservative on readiness.

Recommendations - do it faster next time

  1. Order of operations: mode_prediction.py bind must run after the final capability-layer rewrite of the solver copy, not before. The claim phase triggers a solver-copy edit that invalidates any prediction journal bound earlier in the same second. Rebind the journal immediately before the first assignment_runner ... --rerun.
  2. Empty TT/MM contract is normal, not a bug: when assignment.task.n_questions == 0 and mode_entries[].run_for_time_supported is true, the lane adapter is proven on SS and GM.start will provide the answer_ids. The runner's "D29 live override" takes this path once SS is CORRECT - the first HELD on this message is expected and the next --rerun clears it.
  3. One rerun absorbs a GM TimeoutError: do not panic; the runner never spent a timed variant during the HELD cycles because the gate failed closed. Rerun once per transient error, not more.
  4. Recognize the carrier in one step: peel_any.py <artifact> prints zip(aes, password infected) and the member type table in a single call. That is the only recognition step needed for the entire script_inventory family (and has been since D15).
  5. Mastery bundle shape: MM artifact is a plain ZIP of nested evidence-*.zip files; adapter already recurses into each (solve_fields_ext + peeled lane_peel). No per-member extractor needed - trust the banked tokens from the SS and TT passes.
  6. Novel carrier for Day 29: polyglot-makefile-*.mk was today's new sub-layer (benign-utility, downloader, launcher, staged-chain). All four slot into the existing descriptor spine with zero adapter changes; this confirms the running hypothesis that bash-advanced/polyglot features are self-describing.
  7. GM TimeoutError pattern: the runner's "readiness gate skipped - unreadable: TimeoutError" is survivable; a single --rerun within 60s normally wins. Do not escalate to needuser on a single TimeoutError.

Lessons banked

  • field_formats.confirm_all banked 44 (SS) + 1,833 (TT) + 23,829 (MM) tokens. The MM run essentially populated the lane's token memory for makefile-variant diffs; future days reusing these ids should be first-contact CORRECT.
  • Password infected for AES-encrypted script-lane ZIPs - already standard, confirmed again here.
  • D29 P0-A fail-closed gate: documented. Do not try to pre-stage the per-CID contract; let GM.start fill it after SS is CORRECT.

References

  • Adapter: challenges/Day 29/📜 Make It Make Sense/scripts/script_wip.py (per-CID clone of _lib/autosolve_script.py).
  • Run log: challenges/Day 29/_board/script-analysis-day-29-makefile-variant-diff.autosolve.agent-1.log.
  • Results JSON: challenges/Day 29/_board/script-analysis-day-29-makefile-variant-diff.autosolve.json.