Tactical LogHuntress CTF 2026 // team archive

Days / Day 28 / Web Exploitation / Fail Open Friday

DAY 28//LANE: WEB EXPLOITATION//STATUS: CORRECT x3

Fail Open Friday

Web Exploitation SSTTMM Modes reported CORRECT in this write-up

🌐 Fail Open Friday — Day 24 Web Exploitation

  • Challenge ID: web-exploitation-day-24-fail-open-friday
  • Lane: web-exploitation
  • Station: Homestation (HS Web lane)
  • Solved by: HS (All three modes CORRECT per GM)
  • Writeup salvage source: HS-held notes captured 2026-10-28 after D28 recovery; preserved into the repo so WS + HS both have the method on record.

Results

ModeResultGame Master flag
Single SolveCORRECTTokena689349485af38f2d0811e8d1ce25977
Time TrialCORRECTTokenb3284507e5bd961148bc944dd9239bc7
MasteryCORRECTToken5ee6d6324289fde6a04ce17b5020bfef

All attempts were untimed. Time Trial and Mastery both started with run_for_time=false and no deadline.

Challenge prompt

"The remote policy decoder is part of an access decision. A valid public decision is documented in the application; investigate what happens when that decision cannot be decoded."

There were no downloadable artifacts — each attempt assigned live target(s). SS and TT asked for one proof from a single target; MM asked for proof-1 through proof-5 from five independent targets.

Method summary

Single Solve + Time Trial (same shape)

  1. GET /api/clues on the assigned target. The response identified the feature route and documented the valid public decision shape: remote:<id>:<base64url-JSON> with a documented standard decision.
  2. Sending a MALFORMED value with the documented prefix (e.g. remote:40b12d94f983:not-base64!) to the feature route returned HTTP 200 with effect: "exceptional_condition_fail_open" and a proof_<hex> value.
  3. Submit that proof_<hex> under the exact answer_id returned by GM start (proof-1 in both SS and TT).

Mastery (5 targets, each a different vuln class)

AnswerTarget techniqueProof
proof-1robots.txt → disallowed /snapshots-89561/ → directory listing exposed *.bak JSONProofproof_89f15f7eba154759b1cc003a967d44ea
proof-2Home-page case_session cookie → /api/timeline reveals case 13636 → /api/cases/13636 returns the proofProofproof_8c9d93bc9dd4475b825c933027ea36aa
proof-3Search-param SQL injection ') OR '1'='1'-- → hidden note with proofProofproof_2427ca8bdce6461f9d63a24bedf36a94
proof-4Reader ?doc= traversal → /run/huntress/objective.txt → proofProofproof_d6657ad4fea147c295e2c8c7edc2ea80
proof-5/api/clues → documented remote:6fe40b6a2a03: prefix + malformed base64 → proofProofproof_4db29cadca7e4de5a4ede92ab346818c

Lessons / faster-next-time

  • Each mode's targets are per-attempt. The route suffixes, session cookies, export IDs, and proof values change between assignments. Always use the current attempt's assigned URLs.
  • Follow visible clues first. /api/clues, robots.txt, linked index pages, and the public challenge prompt beat guessing paths.
  • Preserve session cookies. APIs that look standalone (/api/timeline, /api/cases/<id>) may require a home-page session cookie first.
  • For Mastery, inspect each target independently. Mastery can be a MIXED bundle of vuln types (archive leak, IDOR, SQLi, LFI, audit-event fail-open) — do not assume one exploit applies to all.
  • Submit the exact answer_id from the start response (proof-1 … proof-5 here, exact names only).
  • Don't guess answers; derive from live responses. The GM is never a value-probe.

Script

scripts/huntress_fail_open_probes.py — reconstructed probes for SS, TT, and the five Mastery target types. Read-only: prints proof values; does NOT submit to GM. The agent submits exact proofs under exact answer IDs via autosolve.py.

HS lane adoption

HS web lane (web-exploitation) handles this via agent-6/7/8 with CID-hash ownership. The same autosolve_web.py adapter can learn these 5 probe shapes (archive leak / IDOR / SQLi / LFI / audit fail-open) once; TT/MM mutate per attempt but shape is identical.