Queue the Confusion — Day 30 Web Exploitation
Modes CORRECT: Single Solve, Time Trial, Mastery (6 lane-sampler targets).
Prompt
Your job went into the queue with your name on it. Loosen the binding and watch somebody else's outcome come out wearing your stamp.
Single Solve / Time Trial — Pyramid queue owner-binding bypass
The live target is a Pyramid training app:
GET /api/clues->{feature_route, secondary_route, owner_directory_route, current_role: analyst}.GET /api/owners->{caller: analyst, restricted_owner: executive}.POST <feature_route>with{"owner": "executive"}is accepted (server never checks the caller is allowed to queue a job for a different owner) and returns{ok, id, queued_owner: executive}.GET <secondary_route>?id=<id>audits the job. For a job whose caller != owner the audit returns the proof:{ok, effect: "queue_owner_binding_bypassed", evidence: "proof_..."}.
The vulnerability: the queue binding (which owner the job belongs to) is set from the client-supplied owner field without a server-side check against the current role. Submitting owner=<restricted_owner> from the analyst role bypasses the binding; the audit route confirms the mismatch and emits the proof.
Mastery — Lane Sampler (6 different vuln classes, one proof each)
Target 1 — Status Sprout (Node)
- Comment in
/hintsreview crawler policy. /robots.txtdiscloses/snapshots-13fb4/.- Directory listing exposes
deployment-notes-470ca9.json.bakwhose JSON contains the proof undersupport_note.
Target 2 — Case Crate (session + IDOR)
- GET
/sets acase_sessioncookie;/api/historythen returns aevents[]log that referencescase_id: 17196withowner: executive, visibility: reference_only. GET /api/tickets/17196with the session cookie returns the restricted ticket whosenotefield is the proof.
Target 3 — Search Party (boolean-blind SQLi)
/query?q=usesSELECT ... FROM articles WHERE title LIKE '%<q>%'-style concatenation (confirmed by schema extraction below).- UNION is suppressed, but boolean injection works: pivot on
endpoint' OR (<expr>) OR '1'='1(3 cards = TRUE, 1 card = FALSE). - Extracted
CREATE TABLE articles (id INTEGER PRIMARY KEY, title TEXT, topic TEXT, summary TEXT, visibility TEXT)and found a 4th row withvisibility='private'. - Extracted
summaryfor the private row -> proof.
Target 4 — Runbook Nook (PHP LFI)
/reader.php?doc=<path>includes with base/var/www/html/pages/;php://filterprefixes get prepended, so direct wrapper abuse fails, but path traversal works.?doc=../../../run/huntress/objective.txtreturns the proof (depth 3 traversal).
Target 5 — Form and Function (Django owner-binding bypass)
- Only
feature_routeexposed (no secondary audit route). POST /submit-<token>with{"owner": "executive"}returns directly{ok, effect: "framework_authorization_bypassed", evidence: "proof_..."}.
Target 6 — Object Queue (Pyramid, same shape as single/TT)
- Same
api/clues-> owners -> feature POST -> secondary audit chain as the Single, with its own route tokens; proof extracted from the auditevidencefield.
Solvers
scripts/solve_queue.py— generic Pyramid queue-owner-binding handler (SS, TT, T6).scripts/run_timed.py— opens the timed window (timed_start_authorised(1)), reads live targets from GM status, dispatches per family, submits.scripts/t3_blind.py— boolean-blind SQLi extractor keyed on the 3-vs-1 card oracle, binary-searches char codes 32..126.
Lessons + D+1
- The three Pyramid/Django "owner-binding bypass" targets (SS, TT, Target 5, Target 6) are a single family:
GET /api/cluesto learn routes, then POST the restricted owner. Target 5 (Django) returned the proof directly from the feature route; Pyramid variants split intofeature + secondary audit. The Mastery lane-sampler also pulls in robots-backup, session-cookie IDOR via/api/history, PHP LFI with a leading directory prefix, and a boolean-blind-only SQLi (UNION blocked). Keep the solver banked underweb_wip.pyplus a per-family dispatcher. - Stall time: ~15 min, almost entirely on Target 3 (missed that UNION was silently filtered; iterated through MANY union variants before switching to the 3-vs-1 boolean oracle). Lesson: the moment UNION returns zero rows across every column count AND a boolean oracle is already proven, go straight to boolean-blind. The oracle
pivot' OR (<expr>) OR '1'='1with the "3 cards = TRUE, 1 card = FALSE" count is cheap and extremely stable. - Keep in clone:
solve_queue.computeis generic enough across Pyramid/Django queue variants; the Target 3 blind extractor is also reusable for any "/query?q=" style SQLi that has a stable row-count oracle. - Port to shared: generic 3-vs-N boolean oracle helper and a
union_or_blinddispatcher (if UNION returns no rows, auto-fall-back to boolean binary search onsubstr/unicode).
Flag
<REDACTED_TOKEN>
Commander follow-ups (READ THESE)
Three notes dropped in this same folder for the Commander to action BEFORE the next day's release — they are not optional. The Day-30 wall-clock would have been <2 min with these in place; most of the fixes benefit every lane, not just web.
COMMANDER_NOTES.md— web-lane-specific gaps this challenge exposed (silent "autosolve FINISHED" with zero work, opaquetimed_start_authorisedgate, missing lane-sampler banner, seven family handlers to bank inkb/web_exploit_families.json, boolean-blind extractor to lift into_lib/web_kit.py,is_valid_web_proofenforcement).SPEEDUP_PARALLELISM.md— fan-out / recognize-dispatch / adversarial-fork patterns that apply to every lane.assignment_runner.py --fanout N+item_runner.pyworker + "AUTO-FANOUT" claim banner. Est. 10x on Mastery.SUPERFAST_IMPROVEMENTS.md— the broader list (prebrief forks, recognise- in-claim, three-tier model routing, nightly dispatcher regression, handler contract, wall-clock instrumentation, workflow-for-a-day, HELD residue schema, snapshot staleness enforcement). Priority order at the bottom.
Canonical copies also live at docs/commander/SPEEDUP_PARALLELISM.md and docs/commander/SUPERFAST_IMPROVEMENTS.md.