Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Python agentic-aiai-securityalert-triageblue-teamcybersecuritydetection-engineeringincident-responselanggraphllm-agentsmcpmitre-attackmodel-context-protocolpurple-teamsecurity-operationsself-hostedsiemsoarsocthreat-detectionthreat-intelligence
"""The incident report's fallback renderer must not emit untrusted HTML.
`_md_to_html` uses the `markdown` package when it is importable and wraps the
source in `<pre>` when it is not. The fallback interpolated the Markdown raw,
and the title interpolated the case id raw.
Found while fixing GHSA-w754-prh8-m56j, by looking for the same shape
elsewhere. What makes it worth a test rather than a quiet fix is that the
sibling renderer in `app/orchestrator/report.py` already escapes here, and its
comment says why in as many words: "a raw `<pre>` wrap would leak HTML through
unchanged". One copy of the lesson was written down and the other copy did the
thing it warns about.
The content is not trustworthy: the Markdown is model output and the report
embeds connector-supplied entity names, so a hostname or filename carried in
from an alert reaches an analyst's browser.
"""
from __future__ import annotations
import builtins
import re
import pytest
from app.investigator.report_writer_agent import _md_to_html
SCRIPT = "<script>alert('xss')</script>"
IMG = "<img src=x onerror=alert(1)>"
@pytest.fixture
def without_markdown(monkeypatch: pytest.MonkeyPatch):
"""Force the `<pre>` fallback.
The defect is only reachable when `markdown` is absent, which is the
branch a slim image takes — so a test that did not force it would
exercise the safe path and pass against the vulnerable code.
"""
real_import = builtins.__import__
def _refuse_markdown(name: str, *args, **kwargs):
if name == "markdown":
raise ImportError("forced for this test")
return real_import(name, *args, **kwargs)
monkeypatch.setattr(builtins, "__import__", _refuse_markdown)
def _title_of(document: str) -> str:
found = re.search(r"<title>(.*?)</title>", document, re.S)
assert found, "the report has no title element"
return found.group(1)
class TestTheFallbackRendererEscapes:
def test_markdown_body_cannot_carry_an_element(self, without_markdown) -> None:
document = _md_to_html(SCRIPT, "case-1")
assert "<script>alert" not in document
assert "<script>" in document
def test_the_case_id_cannot_carry_an_element(self, without_markdown) -> None:
"""The title took the case id raw.
Asserted structurally rather than by substring: `html.escape` encodes
the angle brackets but leaves `onerror=alert(1)` as literal text, so a
substring check passes on escaped output and proves nothing.
"""
title = _title_of(_md_to_html("# ok", IMG))
assert not re.search(r"<(img|script)", title)
def test_the_document_stays_a_single_document(self, without_markdown) -> None:
"""A payload closing `</html>` and opening another would not show up
in either assertion above."""
document = _md_to_html("</body></html><html><body>grafted", "</title></head><body>grafted")
assert document.count("<html") == 1
assert document.count("</html>") == 1
class TestTheMarkdownPathIsUnchanged:
def test_ordinary_markdown_still_renders(self) -> None:
"""The fix must not reach the `markdown` path, which handles its own
escaping — double-escaping there would corrupt every real report."""
document = _md_to_html("# Heading\n\nSome **bold** text.", "case-1")
assert "Heading" in document
assert "&lt;" not in document