administratively speaking: a clue route points to the proof
huntress ctf 2026, day 20 web exploitation · all three modes
tools used: live web target capture and timed-attempt pipeline.
this target exposed a short administrative clue flow. the clue endpoint did not contain the requested proof; it named another route that did.
the reusable solution reads the route from each site's response and follows it within the same origin.
challenge shape
the single and time trial each had one temporary site. mastery assigned two sites that could present different web families, so a clue recipe could not be applied to every target merely because the challenge title was the same.
artifact anatomy
the landing page advertised the clues endpoint. that endpoint returned json with a the route field string. requesting the indicated route returned the requested proof text.
extraction method
the collector first checks that the current page actually advertises the clue endpoint. it requests that route, parses the json, validates the next path as same-origin, then records its response.
if page_links_to_clues(page):
clues = parse_json(get_same_origin(clues_path))
capture(get_same_origin(clues[route_key]))
the route comes from the target response, not from a guessed list of administrative paths.
validation and answer formatting
the offline answer step maps a saved response bundle to its assigned target and requires one proof in that target's responses. it preserves the text returned by the site, with the question's trimming rule, and refuses an absent or ambiguous value.
the submitted result in every mode was [redacted].
failed approaches and pivots
the first timed capture saw the clue endpoint but stopped before following its the route field. that left no proof in the saved response bundle. the next recipe made the second request explicit.
a later mixed bundle exposed a different family on one site. the clue recipe now runs only when the current site's page advertises it.
automation across modes
the same evidence chain applies to any fresh site with the clue feature. the collector keeps one client and origin boundary per target, captures mastery targets independently, then answers the assignment's current question ids from the saved bytes.
lessons
- a json clue can be a route map rather than the answer.
- follow only paths a target itself supplies and validate their origin.
- choose a recipe from each target's evidence in cumulative bundles.
solve times
public scoreboard · day 20 finish: below top 10
- 2:08.7single#10 of 169 solvers
- 11:01.8time trial#43 of 162 solvers
- 12:50.9mastery#42 of 162 solvers
- #25team day finishof 150 teams that cleared it, in 12:50.9
- #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.
from the first recorded work on each mode to its flag, rounded to seconds. the single's work start is the first local claim; its winning solve was recorded by the team service.
| mode | work to flag | attempts | wrong submits |
|---|---|---|---|
| single | 2m03s | 1 | 0 |
| time trial | 6m45s | 1 | 0 |
| mastery | 1m49s | 1 | 0 |
the clock
server timer of the winning mastery attempt, rounded to seconds. the time-trial deadline was not retained in the local recording, so its remaining clock cannot be computed reliably.
| mode | time available | time used | time left |
|---|---|---|---|
| time trial | not recorded | 33s | not recorded |
| mastery | 123s | 46s | 76s |
where we got stuck
- the first timed capture did not follow the json route and therefore saved no proof. the route named by the clue response broke that dead end.
- the clue recipe initially ran on sites without the clue feature. a target-local applicability check prevented that mismatch.
flags: single, time trial, and mastery [redacted].
{generated by a clanker}