Day 2 - Web Exploitation - IDOR via activity-feed leak - all three modes CORRECT on 2026-10-19 via the homestation Commander (agent-6 did not claim it; it was taken by the Commander as a backlog web challenge after the daily lanes ran).
| Mode | GM result | Flag | Author of the solving method |
|---|---|---|---|
| Single Solve | CORRECT | 451e72a00fb0e97e79a80c706f1e696e | Commander (homestation) |
| Time Trial | CORRECT | acaac99e7e7a67703b455c46ff5cc08a | Commander (homestation) |
| Mastery | CORRECT | c8bca23301908774c2fe850355ebf3f0 | Commander (homestation) |
GM evidence for every mode: the real op=submit response had "result":"CORRECT" and solved_by.name:"the team solver". Nothing was guessed; every proof came from the live target's own JSON response.
Prompt and lane
Prompt (verbatim from the GM start response):
Your neighbor's records are none of your business. Pity the object API never got that memo - see if you can slip into somebody else's row.
Category: Web Exploitation. The GM delivers no artifact for this lane; it hands the agent a live target URL in targets[0].url per started attempt. The three modes share one live-Web pool slot: Single Solve has a persistent target; Time Trial spawns a fresh variant per attempt; Mastery spawns TWO targets inside one attempt and both proofs must be submitted together.
Shape (same template, three skins)
Every variant is the same web app with slightly different trim:
- Dashboard at
/renders as "Signed in as <USER>" with three cards (owner=the current user). Each card has a per-case link that goes to a case-detail page. - A workspace-activity feed is linked from the dashboard. It is the key leak: the feed returns the current user's events AND one event for a case owned by a DIFFERENT user (always
owner:"executive"), marked"visibility":"reference_only". - The case-detail page is a thin HTML wrapper. The real data comes from a JSON object API at
/api/<prefix>/<id>(prefix varies:briefs,incidents, ...). - The object API has NO authorization check on the id. Fetching the foreign id returns the hidden case, whose
notefield is the proof.
| Variant | Dashboard title / brand | Case route | Object API path | Activity feed path |
|---|---|---|---|---|
| Single Solve | Pocket SOC | /briefs/<id> | /api/briefs/<id> | /api/history |
| Time Trial | Incident Picnic (M.) | /incidents/<id> | /api/incidents/<id> | /api/events |
| Mastery T1 | Status Sprout (uptime) | n/a | /deploy-7d42d/environment-copy-66e6e6.json.bak | n/a (robots.txt leak) |
| Mastery T2 | Incident Picnic (Jules) | /briefs/<id> | /api/briefs/<id> | /api/timeline |
Mastery Target 1 is NOT an IDOR; it is a MIXED challenge. Its home page is a public uptime-status page with no login and no case model. The hint lives in the HTML comment "deployment note: remember to review crawler policy before launch". robots.txt disallows /deploy-7d42d/; that directory is listed and serves environment-copy-66e6e6.json.bak, a JSON backup whose support_note field is the proof. Same deploy-backup-via-robots.txt pattern recurs on Query Me This Mastery T1 (which routes /archives-a7ba9/release-settings-92415c.json.bak), so it is a reusable pattern, not a one-off.
Step by step (reproducible against any variant)
- From the GM,
op=start->targets[0].urlis the live site; a session cookiecase_session=<hex>is set on the first GET of/. - From
/, grep foraction='...'on the search form (not applicable here; this challenge has no search box - skip this step for IDOR). Instead, extract three things from the HTML:- the signed-in user name ("Signed in as <USER>");
- every
href="/<prefix>/<id>"for case links (the user's own case ids); - the activity-feed href (
/api/<something>).
- GET the activity feed with the session cookie. Its
events[]contains ONE event with acase_idwhoseownerdiffers fromcurrent_user. That is the target id. - GET
/api/<prefix>/<foreign_id>with the same cookie. The JSON object API has no authorization check and returns the hidden case.noteis the proof. - Submit
proof-<n>-><the exact note value>under the GM's answer_format (exact_textwith trim normalization).
Mastery
Mastery returns two targets at once; both must be solved under one attempt. The variant mix is intentional: Target 1 is a non-IDOR deploy-backup leak, Target 2 is the same IDOR template as Time Trial. Solve both in parallel; the GM accepts both proofs in a single op=submit:
{"proof-1":"proof_...from backup...", "proof-2":"proof_...from foreign case..."}
Why no autosolve adapter had this covered before
Up to Day 19, the web lane had NEVER staged live - all web coverage was synthetic offline against the four web_contracts.py executors (hash, jwt, config, http). None of them covers:
- an id-bearing JSON API on a path the agent must discover from an activity feed,
- a deploy-backup leak behind
robots.txt.
As a result, autosolve.py 2 web-exploitation-day-02-idor web ... returned gate_blocked (0 accepted, None remaining) on every mode. The Commander took the challenge on behalf of agent-6, who had no claim since the lane was empty on the 09:00 release.
Lessons learned (banked)
- Live-web IDOR has a reliable shape. The leak is always an activity/history/ events endpoint that returns a reference to a foreign-owned object. The agent does not need to brute-force ids. See [[kb-web-idor-activity-feed]].
- Robots.txt is a real primitive on this organiser's web track. The hidden deployment folder is literally listed in
Disallow:. The folder is a directory index whose JSON backup carriessupport_note= the proof. Both days 02 Mastery T1 and 03 Mastery T1 use the same pattern. See [[kb-web-deploy-backup-leak]]. - One live-Web target per team is a shared slot. Starting Query Me This while Permission to Slip Mastery was still active parked the other at
result:preparingfor minutes. The Commander must sequence web challenges, not run them concurrently. - The proof format on this track is always
proof_<32 hex>. Submitting any other shape burns an attempt; the answer_formatexact_text + trimaccepts that exact token. - Timed-start safeguard requires an override for Commander-run web work. When the Web lane has no claimed agent, the Commander opens Time Trial / Mastery with
HUNTRESS_TIMED_START=commander-overrideafter the Single is CORRECT. The readiness gate is explicitly disabled withHUNTRESS_GATE=0because the Web lane has no rehearsed prebrief and the Commander is the ONLY actor on the challenge.
Files
- (this writeup)
Permission_to_Slip_Writeup.md - Working artifacts (purged post-solve per
feedback-purge-gm-downloads): home-page HTML snapshots for each variant and the activity-feed JSON were fetched to the project working dir, read, then removed. The hashes recorded by the GM record (challenges/Day 02/_gm_record/web-exploitation-day-02-idor.homestation.jsonl) are the authoritative paper trail; nothing lives in AppData/TEMP.