queue the confusion: an audit that trusts the queue

huntress ctf 2026, day 30 · web exploitation · three modes

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

the target presented itself as a pyramid-based job queue. its clue document named a process route, an audit route and an owner directory. the process route accepted any owner the body named, and the audit route trusted the queued record when it decided what to show.

challenge shape

single and time trial each assigned one temporary target and one answer. mastery assigned multiple targets and answer ids. each assignment rotated the process route, the audit route and the host. the answer for each target was the proof string the audit route returns for a job queued for the restricted owner.

artifact anatomy

the useful artifact was the saved sequence of http responses. the home page linked to a clue document. the clue named the current process route, the secondary audit route and an owner directory route, with the caller's current role. a plain get of the process route answered 405 post_required with request_fields: ["owner"]. the audit route answered 400 naming its own query parameter (job_id_required, query_parameter: id). the owner directory answered 200 with the caller's name and a restricted owner.

extraction method

we read the directory, then queued a job naming the restricted owner. the process route records the owner verbatim — the 201 response echoes queued_owner — and the audit route, keyed by the id it taught us in its own rejection body, shows the record it trusts. auditing the restricted owner's job returns the proof; auditing a job queued for the analyst returns {"status": "empty"}.

validation and answer formatting

each target's proof appears only in that target's audit response. the job ids are per-target and rotate with the assignment, so nothing carries forward between modes.

failed approaches and pivots

the earlier post-then-audit recipe expected an export report with three schema fields — a different audit shape entirely. queueing a job for the caller's own role succeeded but audited empty, because the trust decision only fires on the restricted owner's record. reading the audit route's 400 body for its query parameter name was the unlock: assuming the usual id would have worked on day 29, but only because the target named it too.

automation across modes

the shared collector now prepends explicit owner-swap shapes from disclosed owner scalars to the post-only inspector, collects id-like scalars from each job response, and audits each one under the parameter name the audit route itself discloses, with one loopback-header retry on a denied read. a renamed variant of this class resolves in the first capture.

lessons

solve times

public scoreboard · day 30 finish: below top 10

  • 1:04.2single#6 of 137 solvers
  • 1:11.4time trial#3 of 132 solvers
  • 1:23.1mastery#2 of 116 solvers
  • #14team day finishof 52 teams that cleared it, in 2h43m
  • #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.

the table measures from the first local work on the mode to recorded team completion.

mode work to flag local attempts wrong submits
single about 49s 1 0
time trial about 5s 1 0
mastery about 10s 1 0

the clock

the winning timed attempts' assignment deadlines were not preserved in the local record.

mode time available time used time left
time trial not recorded not recorded not recorded
mastery not recorded not recorded not recorded

where we got stuck

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