Hunt: XSS - postMessage expert (microskill)
Web-message expert of the xss split: the SOURCE is not a URL param but a cross-window message. The attacker page owns the sender; the victim listener owns the sink. The verdict turns on origin verification, and most verifications are string checks - bypassable by construction.
Wiki
Hub: [[wiki/techniques/web/xss]] - qmd on hint only: "postmessage web message origin bypass addEventListener iframe innerHTML eval".
Paths (binding - shared with the mapper run)
RUN=$ENG_DIR/hunt/xss/<host> # SHARED with hunt-xss
Experts write probes/xss-msg*.jsonl into the mapper's $RUN tree.
Step 1 - Listener inventory (JS corpus, read-first)
Full-read the corpus; every addEventListener('message', ...) /
onmessage = handler gets:
handler location (file:line)
what it reads: e.data / e.data.html / nested field paths
which sink receives each field (innerHTML, eval, document.write,
location, el.src, JSON.parse->render)
the origin check (exact code, not a summary)
frames that host the listener (main page vs iframe)
Record @ $RUN/raw/msg-listeners.md rows listener | origin check | data path | sink.
Step 2 - Origin-check verdict
no check ..................... directly exploitable
indexOf/startsWith/endsWith/substring on e.origin
-> bypassable: attacker origin like victim-com.evil.example
(startsWith) or evilvictim-com (endsWith)
regex match .................. check anchors (^https?://exact-host$)
allowlist Set/switch ......... enumerate every allowed origin
e.isTrusted ................. false for postMessage - never a control
Step 3 - Payload delivery (attacker page harness)
<iframe src="https://victim/page"
{html:'<img src=x
- Shape-confused probes: html-bearing objects, arrays, unexpected field types reaching the sink (the CVE-2026-27246 Adobe Connect shape: an html field routed to innerHTML).
- Sink-specific payloads per hunt-xss-dom sink table.
- Harness hosted on an attacker-controlled origin (lab domain / burp collab-hosted page); never a third-party victim.
Step 4 - Validation + exit
-
alert(window.origin)executed in the victim frame; console + screenshot into targets//poc/. - Rows appended to $RUN/candidates.jsonl (class=postmessage);
pointer
XSS-MSG-DONE listeners=<n> bypassed=<n> fired=<n>appended to $ENG_DIR/hunt-xss-.md; Deadends per listener (row names the check that held).
Blocked implies research: before any Deadend is claimable on a held
row, (a) walk the shared arsenal (hunt-xss/references/waf-bypass.md)
rows for that control class, and (b) run one wiki query AND one
mcp__wigolo__search for postMessage origin check bypass <target framework>. Results, or a genuine null, land in coverage-depth.md.
"No research tried" is not a claimable state.
Depth walk (mandatory)
Walk EVERY applicable row: every enumerated message handler against its DOM sinks. Record each
row in $RUN/coverage-depth.md as a table | row | state | evidence |
where state is probed, blocked, or N/A, and evidence is a probe
path or a Deadends.md line. A bare probed with no evidence reference
fails the completion gate.
Validation
Proof = attacker-origin message driving a live sink in the target page (harness page + captured execution/exfil) - a listener missing an origin check but reaching no dangerous sink is a hardening note.
False Positives
- Listener without origin check that only writes to console or benign state.
event.origincheck present but you tested from the same origin (self-XSS, not exploitable).- Sink reachable only from an origin you control but the target frames never embed (no delivery path).
Cross-routing
- hunt-xss (mapper) for the injection-point context; hunt-xss-dom for the sink-side traversal; hunt-ssrf when the handler fetches attacker-supplied URLs.
Attack surface
APPROACH: Enumerate addEventListener('message') handlers in the bundle read-first; trace origin checks against attacker-controllable origins (wildcard, weak regex, no check); validate with a benign marker post before payload chains.
AVOID: A message listener without an origin check is a sink candidate - impact needs the handler writing attacker data into a DOM sink.
Stop conditions
Every listener terminal; every bypass claim cites the exact check code; nothing written into skill dirs.