Severity: LOW
Files:

• plugins/html_to_markdown/html_to_markdown.py (lines 74, 76)
• plugins/robots_license_guard/robots_license_guard.py (line 37)
  Rule: Security - ReDoS

Description

Several plugin regex patterns use constructs that could cause polynomial backtracking on malformed input.

Vulnerable Code

html_to_markdown.py:

# Line 74 - Link extraction
text = re.sub(r"<a[^>]*href=\"([^\"]+)\"[^>]*>(.*?)</a>", ..., text, flags=re.IGNORECASE | re.DOTALL)

# Line 76 - Image extraction
text = re.sub(r"<img[^>]*alt=\"([^\"]*)\"[^>]*src=\"([^\"]+)\"[^>]*>", ..., text, flags=re.IGNORECASE)

robots_license_guard.py:

# Line 37 - Meta tag extraction
META_PATTERN = re.compile(
    r"<meta\s+[^>]*name=\"(?P<name>robots|x-robots-tag|genai|permissions-policy|license)\"[^>]*content=\"(?P<content>[^\"]+)\"[^>]*>",
    re.IGNORECASE,
)

Impact

Lower risk than the validators.py ReDoS because:

• Input is HTML from resource fetches, not direct user validation
• Patterns use bounded character classes [^>]* instead of unbounded .*
• Non-greedy .*? limits worst-case backtracking
• Plugin context means fewer attack vectors

However, a malicious MCP server could return crafted HTML to cause CPU spikes during resource transformation.

Suggested Fix

Use more specific patterns or restructure to avoid multiple [^>]* segments.