The Agent Skills Directory

[SAFE]: The skill serves a legitimate cybersecurity purpose, providing educational content and tools for authorized penetration testing of web applications. It includes appropriate legal disclaimers and warnings regarding unauthorized use.
[COMMAND_EXECUTION]: No shell command execution, subprocess calls, or dangerous system-level operations were found in the provided Python script (scripts/agent.py).
[DATA_EXFILTRATION]: The script interacts with external web applications provided as input parameters to identify vulnerabilities. It does not access sensitive local files (e.g., SSH keys, credentials) or exfiltrate environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external web responses to detect reflections of test canaries.
Ingestion points: Response text from requests.get and requests.post in scripts/agent.py.
Boundary markers: None; the script performs simple string matching to verify payload reflection.
Capability inventory: Network requests via requests library and JSON report generation.
Sanitization: None; however, the script does not execute the response content or pass it to an LLM, mitigating the risk of injection during the automated scan process.
[EXTERNAL_DOWNLOADS]: The documentation references well-known and trusted security resources including OWASP, PortSwigger, and Google's CSP Evaluator. No unauthorized remote scripts or binaries are downloaded or executed by the skill's code.
[OBFUSCATION]: Analysis of the skill body and scripts revealed no hidden content, Base64-encoded payloads, homoglyph attacks, or zero-width characters intended to hide malicious behavior.

testing-for-xss-vulnerabilities