Tactical LogHuntress CTF 2026 // team archive

Days / Day 25 / Web Exploitation / Reset Expectations

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

Reset Expectations

Web Exploitation SSTTMM Modes reported CORRECT in this write-up

Day 25 - Reset Expectations (Web Exploitation)

Agent: agent-7. All three modes CORRECT (Single Solve, Time Trial, Mastery).

Challenge shape

Prompt: "Forgot your password? This recovery flow forgot its manners. Walk the reset process sideways and wake up in somebody else's account."

Single Solve was one Symfony-styled target with a two-step password recovery flow. Time Trial was the same shape on a mutated instance. Mastery bundled FOUR distinct web apps, each with its own vuln class; only Target 4 was the recovery flow.

Single Solve / Time Trial - Account Recovery IDOR

Each target advertises its own routes:

  • GET /api/clues -> {feature_route (submit), secondary_route (result), recovery_guide_route}
  • GET /api/recovery -> {request_route, redeem_route, account_parameter, token_parameter, accounts[]} where accounts is a list of {id, label, self_service} pairs. The self-service account is the attacker's analyst; the non-self-service one is the Executive.

Flow:

  1. GET {request_route}?account_id={self_id} -> JSON {ok, account_id, token, redeem_route}.
  2. GET {redeem_route}?account_id={executive_id}&token={token} -> JSON {ok, effect: "synthetic_account_recovery", account_id: executive_id, evidence: "proof_..."}.

The server validates the self-service gate only on the request step and binds the token to the account that was sent in. The redeem step trusts the account_id in the query and re-derives nothing from the token, so a legitimately minted analyst token lets you redeem into any account.

Single Solve proof: Proofproof_5ad1884a069c45bb9524d4b5fab7e108 Time Trial proof (per-run, auto-solved): recorded in the autosolve record.

Mastery - four independent apps

Each target-N was a fully different app; only routes[].path was advisory metadata (not required to reach the API). Working exploit per target:

Target 1 - Status Sprout (Node.js) - robots.txt leak

GET /robots.txt reveals Disallow: /releases-bdf42/. Directory indexing serves a backup JSON file.

GET /releases-bdf42/environment-copy-ef6f12.json.bak
-> {"environment":"production","support_note":"proof_fc0fcffc5162491fba54eae012ff64ef"}

Target 2 - Tiny Triage - case-id IDOR via activity feed

The home page sets a case_session cookie automatically ("Signed in as Riley"). The owner's three cases are shown, but GET /api/events leaks a fourth case id outside Riley's workspace with visibility: "reference_only" (executive-owner case 13987). The ticket detail endpoint does NOT re-check ownership:

GET /api/tickets/13987
-> {"id":13987,"owner":"executive","title":"Executive escalation","status":"restricted","note":"proof_0e514722a54c4406848d1c60890b3ee5"}

Target 3 - Search Party - SQLi (prefix ) OR 1=1--)

/lookup?q=... returns matching records. Payload ?q=') OR 1=1-- fully escapes the (category=? AND ...) style predicate and lists every record, including the hidden restricted row:

category=restricted, title="Private escalation", body="proof_efe8b1d34473489593c33a35d5a091e5"

Target 4 - Reset Expectations (same shape as Single Solve)

Same two-step recovery flow; the mutated routes still followed the /api/clues -> /api/recovery discovery. Mint a token for the self-service account, then redeem it with the Executive account id. Proof: Proofproof_21e16f1c275146738fdae7b80a52103e.

Submitted answers (GM-accepted)

  • Single Solve proof-1 = Proofproof_5ad1884a069c45bb9524d4b5fab7e108 -> CORRECT
  • Time Trial (ranked, same single-target shape) -> CORRECT by autosolve in 9.0s
  • Mastery
    • proof-1 = Proofproof_fc0fcffc5162491fba54eae012ff64ef
    • proof-2 = Proofproof_0e514722a54c4406848d1c60890b3ee5
    • proof-3 = Proofproof_efe8b1d34473489593c33a35d5a091e5
    • proof-4 = Proofproof_21e16f1c275146738fdae7b80a52103e -> CORRECT

Automation landed

Added _tech_reset_expectations to challenges/_lib/autosolve_web.py so the /api/clues -> /api/recovery -> mint-then-swap chain runs in the shared technique battery for Reset Expectations variants. Also extended _lfi_handler to fold the per-target routes[].path into the base target_url so a Mastery with route-prefixed apps is reachable through the same dispatcher. For the non-Reset Mastery targets the existing battery (robots recon, API clues inventory, SQLi via body scan, etc.) covers the vulns, with the fresh technique as the recovery-specific win.

Notes / lessons

  • Mastery here is a lane sampler, NOT variants of one bug: four different web stacks with distinct vulns on a single timer. Treat each target as a fresh RECOGNIZE/BIND/EXTRACT.
  • A technique that only handles the Single's shape will leave Mastery HELD. Script every target class or let the battery cover them.
  • Session cookies are implicit (set on first /): tooling needs a cookie jar, not stateless curls.