Tactical LogHuntress CTF 2026 // team archive

Days / Day 19 / Web Exploitation / Back Up a Minute

DAY 19//LANE: WEB EXPLOITATION//STATUS: CORRECT x3

Back Up a Minute

Web Exploitation SSTTMM Modes reported CORRECT in this write-up

Challenge

  • Event: Huntress CTF 2026 - Day 1 / web-exploitation lane. Author: John Hammond / Just Hacking Training. Category: Web Exploitation.
  • Modes: Single Solve (2 pts), Time Trial (3 pts), Mastery (2 new dynamic tasks). GM challenge id web-exploitation-day-01-backup-exposure (+/time_trial, /mastery).
  • Prompt: "The deploy left its diary behind. Somewhere out in the open, an old backup is still telling secrets - go find the archive this site forgot to clean up."
  • Access: public GM bearer (https://2026.huntress.ctf.games/), no AI Homebase/VPN/SSH needed. Web Exploitation challenges have no downloadable artifact - start returns live targets[] URLs to open instead.

Triage

  • gm.start(challenge) returns one live target site per question (proof-1 for Single Solve/Time Trial; proof-1+proof-2 for Mastery, one target each).
  • Target site: a small Node.js "Status Sprout" status page. Its HTML carries an inline comment: <!-- deployment note: remember to review crawler policy before launch --> - a direct hint to check robots.txt.

Investigation

  1. GET /robots.txt on the target -> User-agent: * / Disallow: /snapshots-<hex>/. The disallow line names a path that was meant to be hidden from crawlers, not actually access-controlled.
  2. GET that disallowed directory directly -> the server returns a real directory-index HTML page, "Deployment archive - Temporary release files. These should not be public.", linking two leftover files: environment-copy-<hex>.json.bak and release-notes.txt.
  3. GET the .json.bak file -> a small JSON object:
    { "environment": "production", "support_note": "proof_3e5a5a6ef1004eb5a5b45562d4126d9a" }

    release-notes.txt was a red herring ("Deployment completed. Public status checks are green.").

  4. The support_note value is exactly the proof_... string the question's answer_format (exact_text, trim) wants.
  5. Submitted it under proof-1 -> CORRECT. Time Trial and Mastery were started as untimed/regular attempts (no run_for_time) immediately after and solved with the identical method against their own mutated target bytes (Mastery has two independent targets, one proof string each).

Flag Recovery

  • The proof string is whatever value a backup file leaked under the disallowed-but-unprotected directory the site's own robots.txt names. It is deterministic per attempt: robots.txt -> disallowed dir listing -> .bak/leftover file -> proof_<hex> field.

Flags

  • Single Solve: Token2a1b35842913d0e6d9a8d378e663fbda
  • Time Trial (regular/untimed, CourseStack points; ranked attempts left untouched): Token0e701553febccafe4df9ef07f9baf7ce
  • Mastery (2/2 targets, regular/untimed): Token93fe263018dd31027b63d4e2858f3db6

Minimal Reproduction

Against any attempt's target URL T:

  1. GET T/robots.txt -> read the Disallow: path(s).
  2. GET T/<disallowed-path>/ -> directory listing of leftover deploy files.
  3. GET each linked file; the .bak/JSON one contains "...": "proof_<hex>".
  4. Submit that exact string under the question's answer_id.

Automated in scripts/solve_back_up_a_minute.py (fetches robots.txt, follows the disallow path, greps every linked file for proof_[0-9a-f]+, submits). Re-run per mode with the challenge id suffixed /time_trial or /mastery.

Lessons Learned

  • robots.txt Disallow is an exclusion hint for crawlers only, not an access control - always fetch the disallowed path directly; here it served a live, unauthenticated directory index of backup files.
  • Web Exploitation challenges have no static artifact; start hands back a targets[] list of real HTTP(S) origins to probe - treat each as a fresh live recon target, not a downloadable blob.
  • A blind op=start/retry on a /time_trial or /mastery route is refused by ctf_gm.py's no-blind-timed-start guard unless it goes through autosolve.py's readiness gate; once the method is hand-verified against the Single Solve (as here), the named override HUNTRESS_TIMED_START=commander-override lets the Commander proceed without re-deriving a separate gate pass for a one-off, non-day-staged GM challenge outside the lane framework.