edge-special-chars_skill

Overview

This skill verifies safe handling of special characters and common injection patterns when substituting user-provided arguments. It focuses on ensuring that all input substituted into the $ARGUMENTS placeholder is treated as literal text and never executed or interpreted. The skill provides concrete test cases to validate sanitization and escaping behavior.

How this skill works

The skill injects a set of representative strings into the $ARGUMENTS placeholder and observes how the ContentProcessor or agent responds. Test cases include HTML/XML entities, shell metacharacters, path traversal attempts, SQL and command injection patterns, and Unicode/emoji examples. The expected outcome is that none of the inputs are executed, interpreted, or used to break out of the intended content context.

When to use it

• Validating prompt handling in agents that substitute user input into templates.
• Testing sanitization and escaping layers in content processors or template renderers.
• Assessing risk of shell, SQL, or command injection paths in integrations.
• Verifying safe display of user input in HTML/XML contexts and logs.
• Checking path handling to prevent directory traversal vulnerabilities.

Best practices

• Treat all external input as literal text and never pass it directly to shells or interpreters.
• Escape or encode characters appropriate to the target context (HTML, shell, SQL) rather than attempting ad hoc filtering.
• Normalize and validate paths; disallow sequences like ../ that enable traversal.
• Use parameterized queries for database access and avoid string concatenation for SQL.
• Log raw input only when necessary and sanitize logs to avoid terminal or UI control sequences.

Example use cases

• Run the skill against a template renderer to confirm that <, >, &, " and ' are escaped for HTML output.
• Feed shell metacharacters and command patterns to ensure no commands are executed by downstream systems.
• Supply path traversal strings like ../etc/passwd to verify they are rejected or safely normalized.
• Test SQL-like payloads such as ' OR '1'='1 to verify the application uses parameterized queries or rejects the input.
• Validate that Unicode and emoji are preserved as literal characters without triggering processing logic.