resolver revolt: a graph field escaped authorization

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

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

the assigned application exposed a graphql endpoint and a protected object reference. a field resolver returned data for that object even though the ordinary workspace view did not grant access. the challenge lay in finding the current schema and asking for the field that carried the proof.

challenge shape

single and time trial each assigned one target and answer. mastery assigned four target-and-answer pairs, which could mix this graphql application with earlier web families. each target could rotate its endpoint, reference and schema surface.

artifact anatomy

the home page identified the graphql framework and linked a clue document. the clue supplied the current feature route and protected reference. a plain GET of that route returned no object; graphql requests used POST with a json query body. the response separated object data from any resolver errors.

extraction method

the solve used introspection of the current query type, then the object type returned by a field accepting an id argument. the solver requests that field using the protected reference from this target's clues and includes the scalar fields discovered in this target's schema.

query {
  __schema {
    queryType { name }
  }
}

the subsequent query uses the discovered field and argument names rather than a field copied from the single. the proof must appear in the current target's object response before submission.

validation and answer formatting

the route is taken from the current clue response and checked against the assigned origin. the protected reference is used exactly as supplied. the query response must be a successful json object with one proof-bearing target response; an http 200 with graphql errors is not enough. each mastery proof remains paired with the target that produced it.

failed approaches and pivots

the first generic capture requested the feature with GET and received a not-found response. a local single recipe then used a graphql query and won, but its timed successor sent that query to the single's old route. the new target had advertised a different route, so the timed request returned not-found. reading the saved timed capture exposed the stale-route error.

the current shared fallback derives the route and reference anew but still uses the field names observed on day 25. full schema introspection remains the required improvement for variants that change those names.

automation across modes

every target starts with its own home page and clues. the query is built from that target's route, reference and schema. in a mixed mastery bundle, the solver classifies each target separately and maps its proof to the assignment's answer id in target order.

lessons

solve times

public scoreboard · day 25 finish: below top 10

  • 1:43.6single#14 of 155 solvers
  • 3:16.3time trial#20 of 152 solvers
  • 3:47.6mastery#10 of 149 solvers
  • #11team day finishof 138 teams that cleared it, in 6:15.3
  • #6single 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 table measures from the first local work to recorded team completion. the time-trial completion did not retain a winning local capture.

mode work to team completion local attempts wrong submits
single about 101s 1 0
time trial about 92s 1 0
mastery about 18s 1 0

the clock

the time-trial winning assignment was not preserved locally. the mastery row uses its recorded assignment deadline and team completion.

mode time available time used time left
time trial not recorded not recorded not recorded
mastery about 134.9s about 18.2s about 116.7s

where we got stuck

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

{generated by a clanker}