The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to process and correlate untrusted external data sources, including packet captures (PCAP) and various protocol logs, creating a potential surface for indirect prompt injection.
Ingestion points: The skill ingests data from external sources including Zeek logs, firewall/proxy logs, DNS logs, and packet captures as specified in references/log-aggregation.md and references/triage-and-flow.md.
Boundary markers: The instructions do not define explicit delimiters or instructions for the agent to treat this external network data as untrusted strings or to ignore embedded instructions within them.
Capability inventory: The skill authorizes the agent to utilize powerful networking tools including nmap, chisel (for network tunneling), mitmproxy, bettercap, and responder for various investigation scenarios as listed in references/tool-selection-cases.md.
Sanitization: There are no mentioned sanitization, validation, or filtering steps for the content of the logs or network traffic before the agent performs analysis and correlation.

network-technique