The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted network traffic data (such as DNS query names and TLS Server Name Indication values) and incorporates them into a Markdown report. This creates a surface for indirect prompt injection where malicious packet content could attempt to influence the agent's interpretation of the capture. However, the skill is strictly read-only and lacks capabilities for network communication or shell execution, which significantly limits the potential impact.
Ingestion points: The scripts/pcap_analyze.py script extracts strings from packet payloads, specifically DNS query names and TLS SNI extensions.
Boundary markers: The output is structured into a Markdown report with distinct sections, providing clear delimitation for the agent.
Capability inventory: The skill's functionality is limited to reading the provided capture file and writing a Markdown report to stdout or a user-specified file; it contains no subprocess calls or network operations.
Sanitization: Extracted strings are decoded using ascii with errors="ignore", preventing the injection of complex character sets while maintaining basic readability.
[SAFE]: The skill follows security best practices for diagnostic tools, including explicit user confirmation for file paths, a strict read-only policy, and the use of well-known, trusted libraries like Scapy. No malicious patterns such as obfuscation, exfiltration, or unauthorized command execution were detected.

alibabacloud-pcap-analyzer