socket to me: a stream behind a websocket upgrade
huntress ctf 2026, day 27 · web exploitation · all three modes
tools used: live web target capture, saved web inspection, and timed-attempt pipeline.
the target's stream was exposed over websocket rather than ordinary http. the proof could be split across continuation frames.
challenge shape
single and time trial assigned one target each. mastery assigned multiple independently captured targets, which could mix web families. each proof belonged to its own target.
artifact anatomy
the clue supplied a feature route and a stream reference. the endpoint expected a websocket upgrade and a short query parameter carrying that reference; a plain get or the clue field name itself did not identify the stream.
extraction method
derive the route and reference from the current clue, send a same-origin websocket handshake with the stream query value, and validate the upgrade. send a masked client text frame carrying the exact reference in json, then reassemble any fragmented response before looking for a proof.
validation and answer formatting
every route must remain on the assigned origin. current target values are read again in each mode, and a successful status alone is insufficient. a proof is submitted only when the response contains it. submitted results are [redacted].
failed approaches and pivots
the local client initially supported only get and post. its first websocket reader stopped at one final frame and could miss a fragmented message. our own account completed all three modes before our client was ready, and the local work went on for about another minute after the single was solved.
automation across modes
the collector classifies every target independently and keeps each proof paired with its answer id. the same evidence order is applied to each fresh time-trial or mastery target.
lessons
- the upgrade and query alone do not subscribe to the stream; the client must send the reference in a masked frame.
- keep routes, parameters and ids local to the target that supplied them.
solve times
public scoreboard · day 27 finish: below top 10
- 25:55.7single#74 of 151 solvers
- 27:55.7time trial#75 of 147 solvers
- 34:43.1mastery#60 of 135 solvers
- #38team day finishof 127 teams that cleared it, in 34:43.1
- #8single 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 recorded work began with the assigned target. completions by our own account and missing local winning captures are marked as such.
| mode | work to flag | attempts | wrong submits |
|---|---|---|---|
| single | about 26m35s | 1 | 0 recorded |
| time trial | about 8m25s | 1 | 0 recorded |
| mastery | about 30s | 1 | 0 recorded |
the clock
a winning assignment deadline was not retained in the reviewed evidence for each mode; the work-to-flag figures above come from the day timeline where available.
| 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
- the local client initially supported only get and post. its first websocket reader stopped at one final frame and could miss a fragmented message. our own account completed all three modes before our client was ready, and the local work went on for about another minute after the single was solved.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}