Files
hermes-agent/hermes_cli/plugin_validate_desktop.py
teknium1 50e9cd7bd5 fix(plugin-guard): two intake false positives — regex literal <script, allowlist "printenv"
Desktop lint: /<script[\s\S]*?<\/script>/gi in a feed sanitiser scored as
'script injection' and failed pinned-source-validate for rss-reader. Mask
JS regex literals for the markup-shaped rule only; <script in a string
literal (an innerHTML payload) and createElement('script') still fail.

Install scanner: "printenv" as a whole-string entry of a read-only
allowlist (frozenset({..., "printenv"})) fired dump_all_env high →
caution on hermes-jev. Extend the literal-token demotion: a token that is
the ENTIRE quoted literal on a line that executes nothing steps down like
an alternation member; "sudo" inside subprocess.run([...]) and
os.system("printenv") keep high.

A/B vs origin/main: attack probes identical (23 rows), in-tree sweep 319
entries 0 worse/0 changed; both new tests red on base. Bumps
PLUGIN_SCANNER_VERSION to v6 so cached caution verdicts refresh.

Signed-off-by: teknium1 <teknium1@users.noreply.github.com>
2026-09-19 14:30:36 -07:00

77 lines
3.5 KiB
Python

"""Static admission lint for a plugin's Desktop surface (``desktop/plugin.js``).
A ``plugin.js`` is evaluated as ESM in the Electron renderer realm with the app's full authority
(``apps/desktop/src/contrib/runtime-loader.ts`` says so in its header: error isolation only, no
capability boundary). The loader accepts that for files the user put on disk; a catalog install is a
remote source, so listed plugins must stay inside the SDK surface. This lint refuses the moves that
step outside it. It is a tripwire for review, not a sandbox.
"""
from __future__ import annotations
import re
from pathlib import Path
from typing import List, Tuple
# (rule, regex) applied to comment-stripped source; every hit fails the "desktop surface" check.
_FORBIDDEN: Tuple[Tuple[str, "re.Pattern[str]"], ...] = (
("prototype patching",
re.compile(r"\b[A-Za-z_$][\w$]*\.prototype\.[\w$]+\s*=[^=]")),
("prototype patching",
re.compile(r"\bObject\.definePropert(?:y|ies)\(\s*[\w$.]+\.prototype\b")),
("prototype patching",
re.compile(r"\b(?:Reflect|Object)\.setPrototypeOf\(|\.__proto__\s*=")),
("dynamic code evaluation",
re.compile(r"(?<![\w$.])eval\(|\bnew\s+Function\(")),
("dynamic import outside the SDK",
re.compile(r"\bimport\(\s*(?!['\"](?:@hermes/plugin-sdk|react)(?:/[\w/-]*)?['\"]\s*\))")),
("script injection",
re.compile(r"createElement\(\s*['\"]script['\"]\s*\)|<script\b")),
)
_COMMENT = re.compile(r"/\*.*?\*/|(?<![:\w])//[^\n]*", re.S)
# A JS regex literal (``/<script[\s\S]*?<\/script>/gi``) matches markup, it cannot inject any: a
# feed sanitiser that STRIPS script tags is the opposite of the move the rule refuses. Regex
# literals are masked for the markup-shaped rules only; a ``<script`` inside a string literal is
# still the payload of an ``innerHTML`` write and keeps firing. The lookbehind keeps division
# (``a / b / c``) from reading as a literal.
_REGEX_LITERAL = re.compile(r"(?<![\w)\]])/(?:[^/\\\n\[]|\\.|\[(?:[^\]\\\n]|\\.)*\])+/[a-z]*")
_MARKUP_RULES = frozenset({"script injection"})
def _mask_regex_literals(source: str) -> str:
return _REGEX_LITERAL.sub(lambda m: " " * len(m.group(0)), source)
def desktop_surface_findings(source: str) -> List[Tuple[str, int]]:
"""Return ``[(rule, line)]`` for every forbidden construct in a plugin.js source."""
stripped = _COMMENT.sub(lambda m: "\n" * m.group(0).count("\n"), source)
no_regex = _mask_regex_literals(stripped)
findings: List[Tuple[str, int]] = []
for rule, pattern in _FORBIDDEN:
haystack = no_regex if rule in _MARKUP_RULES else stripped
for match in pattern.finditer(haystack):
findings.append((rule, haystack.count("\n", 0, match.start()) + 1))
return sorted(findings, key=lambda f: f[1])
def check_desktop_surface(report, plugin_dir: Path) -> None:
"""Fail the report when ``desktop/*.js`` steps outside the SDK surface; silent when there is none."""
desktop = Path(plugin_dir) / "desktop"
if not desktop.is_dir():
return
hits: List[str] = []
for js in sorted(desktop.rglob("*.js")):
try:
source = js.read_text(encoding="utf-8", errors="replace")
except OSError:
continue
rel = js.relative_to(plugin_dir).as_posix()
hits.extend(f"{rule} ({rel}:{line})" for rule, line in desktop_surface_findings(source))
report.add(
"desktop surface", not hits,
"; ".join(hits[:8]) + (f" (+{len(hits) - 8} more)" if len(hits) > 8 else "")
if hits else "stays inside the plugin SDK surface",
)