element whose only parent chain is
an attribute context — covered by attr-markup; kept as a
separate rule for unquoted-attribute edge cases
Usage:
python3 md_prescan.py < rendered.html
python3 md_prescan.py file.html [file2.html ...]
python3 md_prescan.py --selftest
Output: JSON {"ok": bool, "findings": [{type, line, col, snippet, detail}]}
Exit 0 when clean, 1 when findings exist, 2 on usage error.
Stdlib only. Deterministic — no network, no model calls.
"""
import json
import re
import sys
from html.parser import HTMLParser
MARKUP_TAGS = ("em", "strong", "a", "img", "code", "b", "i", "del", "mark")
class Scanner(HTMLParser):
def __init__(self):
super().__init__(convert_charrefs=False)
self.findings = []
self.anchor_depth = 0
def _add(self, ftype, detail, snippet=""):
line, col = self.getpos()
self.findings.append({
"type": ftype,
"line": line,
"col": col,
"snippet": snippet[:160],
"detail": detail,
})
def handle_starttag(self, tag, attrs):
if tag == "a":
if self.anchor_depth > 0:
self._add("nested-anchor",
" opened inside an open element — a later "
"inline pass linkified text inside an emitted anchor",
self.get_starttag_text() or "")
self.anchor_depth += 1
for name, val in attrs:
if val is None:
continue
hit = re.search(r"<\s*(%s)\b" % "|".join(MARKUP_TAGS), val)
if hit:
self._add("attr-markup",
"attribute %s= contains literal <%s — an inline "
"pass emitted markup inside an emitted attribute"
% (name, hit.group(1)),
self.get_starttag_text() or "")
if "\x00" in val:
self._add("sentinel-leak",
"attribute %s= contains a \\x00 sentinel byte" % name,
self.get_starttag_text() or "")
if name in ("href", "src") and "`" in val:
self._add("attr-backtick",
"%s= contains a backtick — a code span was probably "
"swallowed into the address by a bare-URL pass"
% name,
self.get_starttag_text() or "")
def handle_startendtag(self, tag, attrs):
self.handle_starttag(tag, attrs)
if tag == "a":
self.anchor_depth -= 1
def handle_endtag(self, tag):
if tag == "a" and self.anchor_depth > 0:
self.anchor_depth -= 1
def handle_data(self, data):
if "\x00" in data:
self._add("sentinel-leak",
"\\x00 sentinel byte leaked into rendered text",
data)
def scan(html_text):
"""Return list of findings for rendered HTML text."""
s = Scanner()
s.feed(html_text)
s.close()
# Raw scan for sentinels outside parser callbacks (comments, decls).
for m in re.finditer("\x00", html_text):
line = html_text.count("\n", 0, m.start()) + 1
col = m.start() - (html_text.rfind("\n", 0, m.start()) + 1)
s.findings.append({
"type": "sentinel-leak",
"line": line,
"col": col,
"snippet": html_text[max(0, m.start() - 40):m.start() + 40],
"detail": "\\x00 byte present in raw output",
})
# De-duplicate (type, line, col, snippet-prefix) pairs.
seen, out = set(), []
for f in s.findings:
k = (f["type"], f["line"], f["col"], f["snippet"][:40])
if k not in seen:
seen.add(k)
out.append(f)
return out
SELFTEST = [
# (name, html, expected finding types — empty means clean)
("clean link label url",
'see https://b.example',
[]),
("nested anchor in label",
'see https://b.example
',
["attr-markup"]),
("anchor inside alt",
'hello \x00 world
", ["sentinel-leak"]), ("backtick in href", 'x', ["attr-backtick"]), ("clean emphasis in label", 'a b and c', []), ("clean code after link", 'https://a.example/code',
[]),
("clean img plain alt",
'',
[]),
("nested anchor paragraph",
'',
["nested-anchor"]),
]
def selftest():
ok = True
for name, html_text, expected in SELFTEST:
got = {f["type"] for f in scan(html_text)}
want = set(expected)
if got != want:
ok = False
print("FAIL %-32s want=%s got=%s" % (name, want, got))
else:
print("pass %-32s %s" % (name, sorted(got) or ["clean"]))
print("%d/%d vectors pass" % (
sum(1 for n, h, e in SELFTEST
if {f["type"] for f in scan(h)} == set(e)),
len(SELFTEST)))
return ok
def main(argv):
if "--selftest" in argv:
return 0 if selftest() else 1
texts = []
files = [a for a in argv[1:] if not a.startswith("--")]
if files:
for path in files:
with open(path, "r", encoding="utf-8", errors="replace") as fh:
texts.append((path, fh.read()))
elif not sys.stdin.isatty():
texts.append(("