ask again later: a condition stored for the next read
huntress ctf 2026, day 23 · web exploitation · single, time trial, and mastery
tools used: live web target capture and saved web inspection.
this target accepted an input now and evaluated it when a second route was read later. the response to the write was not the answer; the later result contained the private row.
challenge shape
the site assigned one target in single and time trial, then joined several independent targets in mastery. the generated feature and result paths came from each target's own clues.
artifact anatomy
/api/clues identified Sinatra and provided feature_route and secondary_route. a generic GET of the feature route returned 405 post_required. the result route was a separate read in the same session, after the condition had been stored.
extraction method
send a read-only sql condition through the clue-provided feature route as json. the target accepted the same condition in the candidate input keys. then read its secondary route with the same client and cookie jar.
clues = get_json("/api/clues")
condition = "%' OR 1=1--"
keys = ("query", "filter", "search", "where", "question", "data", "report")
post_json(clues["feature_route"], {key: condition for key in keys})
proof = read_proof(get(clues["secondary_route"]))
validation and answer formatting
require both routes to remain on the assigned origin and require a successful write response before reading the result. copy only the proof in that later response, shown here as [redacted]. the route, session, and proof change with the assigned target.
failed approaches and pivots
the first pass followed the feature route with GET and received the method error. that response explained the method but not the body. a direct search form recipe also missed the delayed result read. the two-step flow matched the clue's paired routes and produced a graded result on a fresh variant.
automation across modes
the collector keeps the write and read inside one target session. mastery targets are classified separately; this condition is sent only when the current target advertises the matching framework and both routes.
lessons
- a successful write can be setup for the real evidence read.
- a
405response can identify the needed method without revealing the request body; combine it with the target's own clue schema.
solve times
public scoreboard · day 23 finish: 1st
- 4:39.5single#13 of 155 solvers
- 5:48.3time trial#9 of 149 solvers
- 7:19.9mastery#9 of 147 solvers
- #1team day finishof 140 teams that cleared it, in 8:58.5
- #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 preserved completion record gives assignment start to flag for the single and mastery. it does not preserve the winning time trial's start or wrong-submit history.
| mode | work to flag | attempts | wrong submits |
|---|---|---|---|
| single | 4m 36.5s from recorded start | 1 recorded | unknown |
| time trial | unknown | unknown | unknown |
| mastery | 68.8s from recorded start | 1 recorded | unknown |
the clock
the challenge listed a nominal 120-second timer for ranked timed attempts. the completion record contains no winning deadline, so the remaining time cannot be verified from it.
| mode | time available | time used | time left |
|---|---|---|---|
| time trial | 120s configured | unknown | unknown |
| mastery | 120s configured | 68.8s from start | about 51.2s if ranked |
where we got stuck
- the first capture saw
405 post_requiredand no proof. following the paired write and result routes with one persistent session resolved that gap on fresh regular variants.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}