Harden report output against stored-XSS via markdown

CHANGELOG.md

@@ -2,6 +2,9 @@
 
 ## [Unreleased]
 
+### 보안
+- 리포트 출력 하드닝 — 생성된 markdown 리포트가 HTML로 렌더될 때 악성 finding 내용(예: 외부 엔진이 스캔한 코드의 `<script>`)이 주입되지 않도록, 프로즈 필드(message/remediation/verification)를 HTML 이스케이프하고 snippet의 code-fence 탈출을 무력화합니다(모든 리포트 타입). rule_id/category/context 등 제약된 식별자는 그대로 둡니다.
+
 ### 추가
 - `appguardrail fix` 명령 — 안전하고 결정적인 자동 수정을 적용합니다(기본 dry-run diff, `--apply`로 기록). 의미를 바꾸지 않는 순수 additive 변환만 수행하며, 첫 변환으로 외부 `target="_blank"` 링크에 `rel="noopener noreferrer"`를 추가합니다(reverse tabnabbing 방지). 동작을 바꾸는 수정(시크릿→env 등)은 위험하므로 자동 적용하지 않고 fix-pack 프롬프트로 남깁니다. scan→fix→verify 루프를 안전하게 닫습니다.
 - `appguardrail sbom` — 의존성 매니페스트(npm `package-lock.json`/`package.json`, Python `requirements.txt`)에서 CycloneDX 1.5 SBOM을 생성합니다. 무의존성(stdlib)으로 동작하며, lockfile이 있으면 resolved 버전을, 없으면 매니페스트의 declared 범위를 사용하고 컴포넌트 properties에 출처를 기록합니다. 공급망 실사(due diligence)의 기본 산출물입니다.

appguardrail_core/reports.py

@@ -69,7 +69,7 @@ def render_buyer_diligence_report(
 ) -> str:
     """Render a buyer-diligence markdown report from normalized findings."""
     context = context or ReportContext()
-    normalized = [normalize_finding(finding) for finding in findings]
+    normalized = [_sanitize_for_markdown(normalize_finding(finding)) for finding in findings]
     normalized.sort(key=finding_sort_key)
     counts = severity_counts(normalized)
     blockers = [finding for finding in normalized if is_deploy_blocking(finding)]
@@ -412,6 +412,40 @@ def _finding_detail(index: int, finding: dict[str, Any]) -> list[str]:
     ]
 
 
+_PROSE_FIELDS = ("message", "remediation", "verification")
+
+
+def _md_prose(value: Any) -> str:
+    """HTML-neutralize free text so a hostile finding message can't inject
+    markup when a generated markdown report is later rendered as HTML."""
+    return (
+        str(value)
+        .replace("&", "&")
+        .replace("<", "&lt;")
+        .replace(">", "&gt;")
+    )
+
+
+def _md_fence(value: Any) -> str:
+    """Prevent a snippet from breaking out of its ``` code fence."""
+    return str(value).replace("```", "`\u200b``")
+
+
+def _sanitize_for_markdown(finding: dict[str, Any]) -> dict[str, Any]:
+    """Return a render-safe copy: prose fields HTML-escaped, snippet fence-safe.
+
+    rule_id/category/context/severity are constrained identifiers used in gate
+    logic and are left untouched.
+    """
+    safe = dict(finding)
+    for field in _PROSE_FIELDS:
+        if field in safe:
+            safe[field] = _md_prose(safe[field])
+    if "snippet" in safe:
+        safe["snippet"] = _md_fence(safe["snippet"])
+    return safe
+
+
 def _short_title(message: str, max_len: int = 84) -> str:
     """Return a compact report heading derived from a finding message."""
     title = message.split(".", 1)[0].strip() or "Security finding"
@@ -432,7 +466,7 @@ def _prepare_report(
 ]:
     """Normalize findings and compute shared report metadata."""
     context = context or ReportContext()
-    normalized = [normalize_finding(finding) for finding in findings]
+    normalized = [_sanitize_for_markdown(normalize_finding(finding)) for finding in findings]
     normalized.sort(key=finding_sort_key)
     counts = severity_counts(normalized)
     blockers = [finding for finding in normalized if is_deploy_blocking(finding)]

tests/test_report_sanitization.py

@@ -0,0 +1,48 @@
+"""Reports must neutralize hostile finding content (stored-XSS via markdown)."""
+
+import pytest
+
+from appguardrail_core.reports import render_report, supported_report_types
+
+HOSTILE = {
+    "severity": "HIGH",
+    "rule_id": "x",
+    "message": "<script>alert(1)</script> <img src=x onerror=alert(2)>",
+    "remediation": "run <b>evil</b>",
+    "verification": "check <iframe>",
+    "file": "a.ts",
+    "line": 1,
+    "context": "app-code",
+    "snippet": "before ``` break out ``` after",
+    "owasp": ["A03:2021"],
+}
+
+
+@pytest.mark.parametrize("report_type", supported_report_types())
+def test_no_raw_html_in_any_report(report_type):
+    report = render_report(report_type, [HOSTILE])
+    # active markup must be escaped, not passed through
+    assert "<script>" not in report
+    assert "onerror=" not in report or "&gt;" in report  # tag broken up
+    assert "<iframe>" not in report
+    assert "<b>evil</b>" not in report
+
+
+def test_escapes_to_entities():
+    report = render_report("buyer-diligence", [HOSTILE])
+    assert "&lt;script&gt;" in report
+    assert "&lt;b&gt;evil&lt;/b&gt;" in report
+
+
+def test_snippet_cannot_break_code_fence():
+    report = render_report("buyer-diligence", [HOSTILE])
+    # the raw triple-backtick from the snippet must be neutralized (zero-width
+    # joiner inserted) so it can't close the ```text fence early
+    assert "``` break out ```" not in report
+
+
+def test_benign_message_unchanged():
+    benign = {**HOSTILE, "message": "Hardcoded secret detected", "remediation": "move to env",
+              "verification": "rerun", "snippet": "const k = 1"}
+    report = render_report("founder-friendly", [benign])
+    assert "Hardcoded secret detected" in report  # no over-escaping