Tactical LogHuntress CTF 2026 // team archive

Days / Day 21 / Web Exploitation / Session Impression

DAY 21//LANE: WEB EXPLOITATION//STATUS: CORRECT x3

Session Impression

Web Exploitation SSTTMM Modes reported CORRECT in this write-up

Session Impression (Day 21 drop, web-exploitation-day-08-session-impression)

GM id: web-exploitation-day-08-session-impression. Name: 🌐 Session Impression. Prompt: "First impressions matter. Plant your session on the welcome mat, then stroll in wearing somebody else's credentials."

Result

All three modes CORRECT (solved_by Blandman3):

ModeResultAccepted answer(s)
SingleCORRECTProofproof_882b49a10a184d1f99b9e56eb7642764
Time TrialCORRECTProofproof_c56c441f3be64ea9b7b0679332fde526
MasteryCORRECT3 proofs across 3 targets (see below)

Flags banked:

  • Single: Tokenbb385c23162649019ea1fe87393b8608
  • Time Trial: Tokenad141b561d077183e422028d2a5fe4a3
  • Mastery: Tokenb6ffd508eeed340b139ec7256a06ee56

Attack

Each Session Impression target is a Laravel training app with the same shape. The attack has three HTTP requests per target:

  1. Recon. GET /api/clues returns the random per-attempt names of the feature/secondary routes, the current role (analyst), and a session_hint string.
  2. Plant. GET <feature_route>?sid=<session_hint> returns {"ok":true,"session_established":true} and sets arena_session=<session_hint> as an HttpOnly cookie. The feature route accepts the sid parameter name only (session_id, sess, sessionId, X-headers, Authorization: all return session_id_required).
  3. Collect. GET <secondary_route> with Cookie: arena_session=<session_hint>. The server has, in the background, logged in a reviewer against the planted session, so the request now returns {"ok":true,"effect":"fixed_session_privilege_transfer","evidence":"proof_<hex>"}. That evidence string is the GM proof.

Mastery delivers three different targets per attempt, each a different vuln pattern:

  • Target 1 — Tiny Uptime (Node.js backup exposure). /robots.txt disallows /releases-<slug>/ or /deploy-<slug>/. That directory lists a *-config-*.json.bak whose support_note field is the proof.
  • Target 2 — Incident Picnic (FastAPI-style IDOR). Signed in as Jules; /api/events includes a case_id owned by executive (visibility reference_only). GET /api/incidents/<that_id> with the auto-assigned case_session cookie returns the executive's case; the note is the proof.
  • Target 3 — Cookie Desk (Laravel session fixation). Same three-step attack as Target 1 of the Single — recon /api/clues, plant <feature_route>?sid=<hint>, collect <secondary_route> with the planted arena_session cookie.

Solver

scratchpad/si_solve.py runs the three-step attack for every target returned by the Session Impression start, maps answer_ids proof-N to targets by order, and submits in one batch. It worked for Single and Time Trial unmodified. Mastery is one target per set and the targets vary by vulnerability pattern, so the three were solved by hand against the same request contract (one curl chain per target).

Why this challenge got put off

It's a web challenge, and web is one of the lanes with no banked autosolver on this machine — it's agent-/Commander-solved live. My first pass today missed it because an unqualified gm.list() without a day= filter returned a filtered subset that happened to omit the three day-07/day-08 web entries served up for Day 21. A gm.list(day=21) returned them. Fix: when enumerating "what's open now", page the GM by day and union the results. Recorded as a lesson so it doesn't bite again.

Files

  • scratchpad/si_solve.py — three-step solver (recon, plant, collect) that works against any Session Impression Laravel target.
  • GM record: challenges/Day 08/_gm_record/web-exploitation-day-08-session-impression.workstation.jsonl.