protocol-state-exploitation

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze potentially untrusted network protocol data and external web content (via WebFetch and WebSearch) while possessing high-privilege tools like Bash, Write, and Edit. It lacks specific defensive instructions or boundary markers to prevent the agent from acting on malicious commands that could be hidden within the captured traffic or fetched documents.\n
  • Ingestion points: External data sources accessed via WebFetch, WebSearch, and local files read through the Read tool (specifically ssh_capture.pcap, tls_capture.pcap, and http2_capture.pcap).\n
  • Boundary markers: Absent. The instructions do not define delimiters or specific "ignore embedded instructions" warnings for the agent when processing protocol data.\n
  • Capability inventory: Bash (shell execution), Write (file creation), Edit (file modification).\n
  • Sanitization: No sanitization logic is described to filter out potential prompt injection payloads from network captures before analysis.\n- [EXTERNAL_DOWNLOADS]: The documentation in test-cases.md and guides/http2-rapid-reset-guide.md recommends installing the h2 Python library using pip3 install without specifying a version. This is a best-practice violation that could lead to non-deterministic behavior or the installation of an incompatible version in the agent's environment.\n- [COMMAND_EXECUTION]: The skill provides numerous command-line examples for security tools such as tshark, tcpdump, nmap, and hping3. While appropriate for a protocol exploitation skill, these enable the agent to perform active network reconnaissance and attacks. The "reverse shell" pattern flagged by static analysis in test-cases.md is confirmed as a false positive; the snippet python3 -c "import socket; s=socket.socket(); s.connect(('target', 22))" is a standard way to open a TCP socket for protocol state testing rather than an attempt to spawn a shell.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM