forged with care: a preference form that reveals target evidence

huntress ctf 2026, day 22 · web exploitation · single, time trial, and mastery

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

this challenge assigns a temporary web application and asks for a proof from its own evidence. the important work is to follow the form and response flow advertised by each target while keeping each target's state separate.

the key insight is that changing a preference is only the first step. the target's response exposes a second, same-origin evidence resource; the proof comes from that response rather than from the form page.

challenge shape

the single and time trial use one live target. mastery supplies three targets, and each target must be analyzed independently because their proof sources can differ.

artifact anatomy

the first page contains a preference form with a destination selector and a hidden value. its action and method describe the required request. a later same-origin response contains the evidence for that target. the mastery set also includes targets whose proof is available through other known evidence flows, so a bundle position alone does not identify the right response.

extraction method

read the form's action, method, hidden controls, and selected option from the captured page. submit those values with the same target's client, then follow the evidence link returned by the server. the collector preserves the target's cookies and checks every requested path against the assigned origin.

form = read_observed_form(capture.first_response)
capture.request(
    form.method,
    form.action,
    data={**form.hidden_controls, form.selector: form.selected_value},
    headers=target_context_headers(capture.origin),
)
capture.get(form.evidence_path)

validation and answer formatting

associate each captured response with its target before matching it to the question. accept one unambiguous proof from that target's evidence; do not combine values from different targets. the graded proof is [redacted] and is copied from the response without normalization.

failed approaches and pivots

the first generic pass treated every preference action like a POST and looked for evidence in the initial clue responses. that missed the form's captured method and the later evidence resource. reading the form controls and following the response's link fixed the flow.

automation across modes

one target-local recipe handles this preference flow, while the solver keeps the other mastery targets on their own evidence recipes. the live collector captures targets concurrently but returns them in assignment order; the offline answer step still uses the target-to-question map.

lessons

solve times

public scoreboard · day 22 finish: below top 10

  • 5:20.8single#51 of 163 solvers
  • 7:18.9time trial#51 of 158 solvers
  • 10:03.4mastery#48 of 155 solvers
  • #26team day finishof 145 teams that cleared it, in 10:39.5
  • #6single orderof the day's 7, by our flag time

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

times are from the first target work to the accepted result. the single uses the recorded release attempt; the timed rows use clean fresh assignments after the capture recipe was corrected.

mode work to flag attempts wrong submits
single 5m 02.6s 1 0
time trial 7.45s 1 0
mastery 9.62s 1 0

the clock

the saved started_at and deadline records give each timed attempt's available window. used time is the clean attempt's start-to-flag duration; time left is calculated from the recorded deadline.

mode time available time used time left
time trial 123.6s 7.45s 116.1s
mastery 124.4s 9.62s 114.8s

where we got stuck

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

{generated by a clanker}