reset expectations: a token crossed an account boundary

huntress ctf 2026, day 25 · web exploitation · three modes

tools used: live web target capture and timed-attempt pipeline.

the recovery flow issued a token to an account allowed to reset itself. the same token could then be redeemed for a different account whose record did not allow self-service recovery. the current target's guide supplied every route, parameter name and account id needed to demonstrate that mismatch.

challenge shape

single and time trial each had one assigned target and answer. mastery contained four target-and-answer pairs, potentially from different web families. new assignments rotated target hosts, routes, accounts and tokens.

artifact anatomy

the home page led to /api/clues. the clue response named a recovery guide. that guide identified the request and redeem routes, the account and token parameter names, and account records with a self_service property. the token came from the request response, not from the guide or another target.

extraction method

we read the current guide and selected one listed self-service account and one listed protected account. the first request obtained a token for the self-service account. the second used the unchanged token with the protected account on the guide's redeem route.

guide = get_same_origin(clues["recovery_guide_route"])
source = next(a for a in guide["accounts"] if a["self_service"] is True)
dest = next(a for a in guide["accounts"] if a["self_service"] is False)
issued = get_same_origin(
    guide["request_route"],
    params={guide["account_parameter"]: source["id"]},
)
result = get_same_origin(
    guide["redeem_route"],
    params={
        guide["account_parameter"]: dest["id"],
        guide["token_parameter"]: issued[guide["token_parameter"]],
    },
)

the proof was read from the redeem response associated with that target.

validation and answer formatting

the solver validates that the guide is a successful json object, both routes stay on the assigned origin, the parameter names are bounded identifiers, and the two accounts have different self_service values. a missing token ends the flow. only a proof present in the redeem response is submitted, and mastery maps it to the answer id for that target.

failed approaches and pivots

the generic first capture followed the clue link but had no recipe for the recovery guide, so it stopped without proof. reading the guide exposed the account distinction and the request order. copying a token or account id from the single would fail on the rotated time-trial target.

automation across modes

the local solver used fresh guide values to win the time trial and mastery. the shared collector now follows the same evidence-gated flow. in mixed mastery it classifies each target independently, so a recovery target does not impose this recipe on its neighbors.

lessons

solve times

public scoreboard · day 25 finish: below top 10

  • 1:54.8single#17 of 157 solvers
  • 2:11.0time trial#9 of 153 solvers
  • 2:20.6mastery#6 of 150 solvers
  • #11team day finishof 138 teams that cleared it, in 6:15.3
  • #7single orderof the day's 8, by our flag time

from the public scoreboard: 09:00 et release to our flag; the table below counts from our attempt's start.

the single's team completion had no local winning capture. the timed rows use the recorded assignment start and completion times.

mode work to team completion local attempts wrong submits
single about 112s 1 0
time trial about 3.6s 1 0
mastery about 8.4s 1 0

the clock

the server deadlines were recorded for the winning local timed runs. values are rounded to a tenth of a second.

mode time available time used time left
time trial 122.0s 3.6s 118.4s
mastery 124.3s 8.4s 115.9s

where we got stuck

flags: single, time trial, and mastery [redacted].

{generated by a clanker}