tcp-ip

Pass

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: SAFEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The C code example for packet analysis (Example 4) utilizes raw sockets via socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL)). This operation typically requires root or administrative privileges (e.g., sudo or CAP_NET_RAW) to execute, as it allows the application to bypass standard protocol stacks and intercept all network traffic on an interface.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides multiple implementations of network servers (Go, Python, and Rust) that receive and process data from arbitrary network clients, presenting a surface for untrusted instructions to enter the agent's context.
  • Ingestion points: Network read operations in SKILL.md Examples 1 (reader.ReadString), 2 (self.socket.recvfrom), and 5 (stream.read).
  • Boundary markers: Absent. The examples directly process, log, and echo data received from the network without delimiters or warnings to the processing agent.
  • Capability inventory: The skill enables the agent to perform bidirectional network communication and potentially execute low-level packet analysis.
  • Sanitization: Minimal. The examples perform basic UTF-8 decoding and string trimming but do not validate or sanitize the content of the network messages before logging or echoing them.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 16, 2026, 06:25 PM