Tactical LogHuntress CTF 2026 // team archive

Days / Day 25 / Web Exploitation / Shell Company

DAY 25//LANE: WEB EXPLOITATION//STATUS: CORRECT x3

Shell Company

Web Exploitation SSTTMM Modes reported CORRECT in this write-up

Shell Company (web-exploitation-day-16-shell-company) - Day 25

Lane: web-exploitation. Agent: agent-6 (Day 24 session, operating on Day 25 pool per user override).

Status - ALL MODES CORRECT

  • Single Solve: CORRECT (untimed). Flag Token48c71e422550552ddbd3785763e24f9b.
  • Time Trial: CORRECT (RANKED, 7.3s via autosolve_web). Flag Tokene3dada4c95e85b79bf29121e4624682e.
  • Mastery: CORRECT (untimed continuation of the active attempt). Flag Token8aad2c3b5e8c645da7b0a419bb2575da.

Single Solve - OS command injection (Gin/Go)

/api/clues advertises feature_route (e.g. /process-38db5f), input_parameter (path), example_path (/app/static). The parameter is passed straight into a shell command (ls -ld $path), so ;/&&/| inject. Running as uid 10001 (arena); the proof is in /run/huntress/objective.txt (env OBJECTIVE_PATH).

GET /process-38db5f?path=/app/static;%20cat%20/run/huntress/objective.txt
# -> {"effect":"worker_command_completed","ok":true,"stdout":"... proof_<32hex>"}

proof-1 = Proofproof_a7b13a4af17d4e55996d4dee0e0315bb

Mastery - 4 targets, 4 distinct web-exploitation classes

Bundle of four independent live targets under one attempt. Derived each proof from that attempt's own bytes:

Target 1 - Green Light (Node.js) - exposed backup via robots.txt /robots.txt Disallow /archives-fa2ae/ -> directory index -> environment-copy-b29d58.json.bak contains the proof in support_note. proof-1 = Proofproof_efc23ea631924f5a8865bc7a9cb13393

Target 2 - Pocket SOC - IDOR Visiting / (or any /briefs/N) sets an implicit case_session cookie. /api/activity leaks an event for a case owned by executive with visibility:reference_only (id outside the current user's workspace, e.g. 17329). /api/briefs/17329 re-trusts the id and returns {"status":"restricted","note":"proof_..."}. proof-2 = Proofproof_3ffa9c074b3b4dcbbc48b5e39cbc0a6c

Target 3 - Archive Atlas - SQL injection (SQLite), blind GET /search?q= concatenates into ... LIKE '%<q>%' on table articles(id,title,topic,summary,visibility). UNION/-- are effectively unusable (comment stripped; UNION yields errors), but quote-balanced boolean injection works. Oracle: persistence%' AND (<cond>) OR '1'='2 -> the "persistence" card appears iff <cond> is true. The search view only returns visibility='public' rows (3 of them); the 4th row is non-public and its summary starts with proof_. Blind-extracted it char-by-char (unicode(substr(...,i,1)) binary search). proof-3 = Proofproof_bb6144aaec9842ec964ef8a90f71fadd

Target 4 - Shell Company - OS command injection (same as Single) proof-4 = Proofproof_a68d9db87f7a4c2d90de274a69c0a8d7

Solvers banked

challenges/_lib/autosolve_web.py technique battery now includes:

  • _tech_shell_company (cmdi via /api/clues feature_route + input_parameter)
  • _tech_search_sqli, _tech_session_idor (added alongside; the SQLi hidden-row case was finished here by blind extraction)

Scripts: challenges/Day 25/Shell Company/scripts/{probe.sh,inject.sh,sqli.sh,blind.py,test_cmdi.py,submit_mastery.py}.

Flags (GM-accepted, not redacted)

  • Single: Token48c71e422550552ddbd3785763e24f9b
  • Time Trial: Tokene3dada4c95e85b79bf29121e4624682e
  • Mastery: Token8aad2c3b5e8c645da7b0a419bb2575da