Challenge: Web Exploitation · GM ID web-exploitation-day-10-uploading-expectations Run: 2026-10-22 Result: Single Solve, Time Trial, and Mastery all returned CORRECT from the Game Master.
Approach
The GM supplies live targets in each mode’s start response. Each target was handled independently; the Mastery targets used three different web paths despite sharing the same challenge prompt.
Single Solve
The homepage exposed upload route /inspect-e628af; /api/clues documented renderer /result-824c5b. The page claimed uploads should be .txt with text/plain, but accepted filename=proof.html with the declared content type still set to text/plain. The uploaded body {{objective}} was evaluated by the HTML preview. The resulting exact proof was accepted under proof-1.
Proof: Proofproof_c999ebff4850488d90d8ed4adb6e3c20
Time Trial
The assigned Time Trial target was a new instance. Its exposed routes were /action-ef679c and /internal-a0d2c4. Repeating the same policy-boundary check on that target produced a distinct proof, accepted under its exact GM answer ID. The route was opened with run_for_time omitted, so this was regular untimed play.
Proof: Proofproof_cf908a6342454bf8b74bacb882b1ec1f
Mastery
The GM assigned three independent targets and answer IDs:
| Answer ID | Target evidence | Proof |
|---|---|---|
proof-1 | robots.txt named /deploy-cecb7/; its directory index linked site-config-3d26c8.json.bak, which contained the proof. | Proofproof_82d7da8539a44935bc7a967e722d641d |
proof-2 | The homepage established Riley’s session. /api/timeline referenced executive-owned case 15076 with reference_only visibility. The visible case pages showed the API route pattern; /api/tickets/15076 returned the proof. | Proofproof_b44d0f14613c40f6852e366e312f5258 |
proof-3 | /api/clues identified upload route /execute-c59588 and renderer /account-896b55. Uploading {{objective}} as objective-preview.html with declared text/plain returned an HTML preview that evaluated the field. | Proofproof_b345dc065b06443684c2c23eebbfff24 |
All three values were submitted together under the exact answer IDs from the Mastery start response; the GM returned CORRECT.
GM-accepted flags
- Single Solve: Token4caf165acb6e00ad23d339b8be1e40f9
- Time Trial: Token971273d0921592534e04806910db6378
- Mastery: Tokend0791bd70d5f12d6bafbbd618f80567a
Reproduction
The attempt-aware solver is solve_uploading_expectations.py. It takes each mode’s own target URLs from the GM start response, bounds HTTP requests to each exact GM-authorized origin, derives each answer from that target, and submits only after covering the returned answer IDs. The original challenge targets expire after the attempt; use a fresh GM attempt’s URLs when reproducing the method.
Per-target evidence is in output/; the GM exchange record is in ../../Day 10/_gm_record/web-exploitation-day-10-uploading-expectations.homestation.jsonl.
Commander handoff and speed recommendations: HOMESTATION_CLAUDE_COMMANDER_HANDOFF.md.
Failed paths and lessons
- Treating every Mastery target as the same upload app failed when Target 1 presented a status page and Target 2 presented a case dashboard. The solver stopped before submitting; inspecting each target’s own links and notes revealed the distinct intended paths. Retry this approach whenever a mode assigns heterogeneous target sites.
- Calling
/api/timelinewithout retaining the session cookie from the homepage returned401 session_required. Fetch the homepage first and keep its cookie only for requests to that same GM target origin. - A previous evidence writer saved all three Mastery records to one filename, leaving only the last entry. The evidence writer now names each target separately; the accepted answers remain independently preserved in the GM record.