the api has opinions: inventorying a target's own request contract
huntress ctf 2026, day 24 · web exploitation · all three modes
tools used: live web target capture and timed-attempt pipeline.
an api puzzle can expose enough of its own contract to answer a protected reference without guessing endpoint names or request fields. the useful work is to follow the current target's inventory and verify the proof-bearing response.
challenge shape
the challenge appeared as one assigned target in single and as a mixed application bundle in mastery. the bundle combined backup, case-portal, and api-inventory behavior. a challenge title was not enough to choose a recipe for each target.
artifact anatomy
the target's clue document named an inventory route, a feature route, and a protected reference. the inventory response described the feature method, accepted query parameters, available fields, and actions. those response bodies were saved as the evidence artifact for offline answer extraction.
extraction method
first read clues from the assigned origin. fetch the inventory route from those clues and confirm its feature route agrees with the clue document. then construct requests only from the reference and the field/action choices the inventory disclosed. one pair returned proof data in this run:
GET <feature-route>?id=<protected-reference>&fields=[redacted]&action=[redacted]
the solver tries the observed pair first and keeps any fallback inside the advertised combinations and a fixed request cap. it extracts proof text from the response body rather than treating an http status as proof.
params = {
"id": protected_reference,
"fields": "[redacted]",
"action": "[redacted]",
}
response = client.get(feature_route, params=params)
validation and answer formatting
the inventory must be a successful json object, name the same feature route, and expose the query parameters, fields, and actions used by the request. the route must remain on the assigned origin. the extracted proof must be present in that target's response and map to the answer id paired with it in the assignment.
for bundles, target order is preserved. each target gets its own session and clue pass, and a known proof can be submitted while other targets are still being investigated.
failed approaches and pivots
an early timed capture reused a feature route from another mode. the new target returned no proof for that route. reading the current target's inventory revealed the actual route and prevented the collector from relying on stale cross-mode state. a former mastery hook also assumed response rows were objects with a path attribute; the saved capture uses dictionary rows, so extraction now follows the actual row schema.
automation across modes
the same evidence sequence applies to single, time trial, and the matching mastery target. route names and references are per-assignment data. the collector captures each target independently, and the offline solver maps proofs through the assignment's ordered answer list.
lessons
- treat the inventory as a request contract and derive every request from the current target's response.
- a response status does not replace finding proof text in the body.
- preserve target and answer order in mixed bundles.
solve times
public scoreboard · day 24 finish: top 3
- 1:18.1single#6 of 159 solvers
- 2:31.1time trial#8 of 156 solvers
- 3:47.1mastery#11 of 153 solvers
- #3team day finishof 139 teams that cleared it, in 3:47.1
- #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 start to the recorded team completion. the timed modes did not retain a successful local proof capture, so these figures are not winning-attempt runtimes.
| mode | work to team completion | attempts | wrong submits |
|---|---|---|---|
| single | 76.6s | 1 | 0 |
| time trial | 87.9s | 1 | 0 |
| mastery | 61.1s | 1 | 0 |
the clock
no successful local timed attempt was retained for either mode, so its winning-attempt deadline and remaining time are unavailable.
| 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 timed recipe carried a route across assignments and spent a capture without proof. reading the current inventory exposed the mismatch.
- a mastery-specific hook read fields absent from the stored row schema and crashed before proof extraction. removing it let the shared parser handle the capture format consistently.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}