The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The deployment workflow requires elevated permissions for multiple system-level operations.
Evidence: Extensive use of sudo in SKILL.md for tasks including package installation, source compilation (make install), network interface tuning (ethtool), and service management.
[PERSISTENCE_MECHANISMS]: The skill establishes automated long-term operation through system scheduling and background services.
Evidence: Instructions in SKILL.md create a daily cron job at /etc/cron.d/zeek-logrotate and enable the filebeat systemd service for log management.
[EXTERNAL_DOWNLOADS]: Fetches software components from an external repository during the installation process.
Evidence: Downloads the Zeek source code from its official GitHub repository (https://github.com/zeek/zeek).
[COMMAND_EXECUTION]: The included analysis script executes system commands to interact with the network monitoring environment.
Evidence: scripts/agent.py utilizes subprocess.run() to invoke the zeek and zeekctl binaries for status checks and PCAP processing.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from network logs, creating a surface for potential injection attacks if the output is processed by an LLM.
Ingestion points: scripts/agent.py reads and parses conn.log, dns.log, and ssl.log, which are populated with raw network metadata such as SNI, DNS queries, and User-Agents.
Boundary markers: No delimiters or explicit instructions are provided to the agent to treat log data as untrusted content.
Capability inventory: The agent has the ability to execute system commands via subprocess.run() and read files from the filesystem.
Sanitization: Network identifiers and queries are parsed into summaries without evidence of validation or sanitization against malicious strings.

detecting-network-anomalies-with-zeek