permission to slip: reference leakage into an idor

huntress ctf 2026, day 19 web track · web exploitation · all three modes

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

the application opened as an authenticated case dashboard. the assigned user could read its own cases, but an activity feed named an executive-owned case that the object api returned without an ownership check.

the key insight was that this was not an id-enumeration problem. the application itself disclosed the interesting object id in a reference-only event; the authorization failure happened on the subsequent object request.

challenge shape

the single and time trial each assigned one temporary target. mastery combined one target from the earlier backup family with one target from this family. endpoint names and object ids varied, so the solution learned them from each target's page and feed.

artifact anatomy

loading the home page established a target-local session cookie and linked the current user's ordinary cases. the page or one of its same-origin scripts also named an activity endpoint. depending on the variant, the route used words such as history, events or timeline.

the feed returned json events. one event referred to an object owned by a more privileged user and marked it as reference-only. an ordinary linked case revealed the object route's shape. substituting the referenced id returned the restricted record and [redacted].

extraction method

the solver retained one http session across the complete chain:

GET /
GET <activity endpoint>
GET <object route>/<referenced id>

it gathers api routes from the page and at most two same-origin scripts, ranks activity-like paths, parses their json, and puts ids from reference-only objects ahead of ordinary ids:

for event in walk(feed_json):
    bucket = referenced if event.get("visibility") == "reference_only" else ordinary
    for key in ("case_id", "brief_id", "id"):
        if str(event.get(key, "")).isdigit():
            bucket.append(str(event[key]))

the object route is learned from an ordinary numeric link when available. a small family fallback handles the observed brief route, but the candidate id still comes from the target's own feed.

validation and answer formatting

the session cookie had to come from the same assigned origin. a stateless feed request returned 401, which distinguished missing session state from a bad endpoint. every followed script, feed and object url passed the origin check.

the restricted response had to contain exactly one requested value. the submitted result in every mode was [redacted].

failed approaches and pivots

automation across modes

recipe selection happened per target, not per challenge bundle. that allowed a heterogeneous mastery bundle to run the backup recipe on one origin and the idor recipe on the other, concurrently and with separate cookies.

lessons

solve times

public scoreboard · day 19 finish: top 3

  • 2:03.0single#2 of 178 solvers
  • 3:18.7time trial#2 of 168 solvers
  • 17:33.0mastery#5 of 168 solvers
  • #3team day finishof 153 teams that cleared it, in 17:56.2
  • #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.

from when we started working on the mode to its flag.

mode work to flag attempts wrong submits
single 2m02s 1 0
time trial 15s 2 1
mastery 24s 1 0

the clock

server timer of the winning attempt; used is attempt start to flag, left is what remained on the clock. the accepted mastery run was untimed.

mode time available time used time left
time trial 120s 15s 105s

where we got stuck

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

{generated by a clanker}