The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions in SKILL.md require the use of sudo to install the Tetragon and Tetra binaries to /usr/local/bin/ and to run the Tetragon daemon. These elevated privileges are necessary and expected for loading eBPF programs into the Linux kernel.
[EXTERNAL_DOWNLOADS]: SKILL.md and references/api-reference.md provide instructions to download Tetragon releases directly from the official Cilium GitHub repository (github.com/cilium/tetragon). This is a well-known service and an appropriate source for the software.
[COMMAND_EXECUTION]: The scripts/agent.py file uses the subprocess module to programmatically execute kubectl, helm, and tetra commands for cluster management, deployment, and event streaming. The commands are constructed using lists, which follows security best practices to prevent shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection as it processes external data.
Ingestion points: The parse_tetragon_log function in scripts/agent.py reads and parses JSON log files generated by the Tetragon daemon.
Boundary markers: None identified in the parsing logic; the script treats all lines in the log as trusted JSON input.
Capability inventory: The agent possesses the capability to execute system commands via subprocess (e.g., kubectl, helm).
Sanitization: The script performs basic JSON parsing and field extraction (binary paths, function names) without explicit sanitization or validation against a strict schema. While the risk is low, malicious processes on a monitored system could potentially craft execution patterns that generate misleading logs to influence the agent's behavior.

implementing-ebpf-security-monitoring