The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download a script from install.crowdsec.net and pipe it directly into a root shell (curl | sudo sh) for installation. This is a high-risk pattern as it bypasses local security verification of the remote content.
[EXTERNAL_DOWNLOADS]: The skill configuration for Traefik references a community-maintained plugin from a third-party GitHub repository (github.com/maxlerebourg/crowdsec-bouncer-traefik-plugin) which is not an official or well-known organization. It also fetches official installation resources from install.crowdsec.net.
[PRIVILEGE_ESCALATION]: Setup and management of the CrowdSec engine require extensive use of sudo, granting the agent root-level access to the system for tasks like installing packages, managing services, and modifying network security rules.
[COMMAND_EXECUTION]: The skill relies on the execution of multiple system-level shell commands to manage bouncers, alerts, and decisions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of system and service logs, as well as HTTP requests via the WAF component. 1. Ingestion points: Monitors system logs (/var/log/auth.log), web server logs (/var/log/nginx/*.log), and real-time HTTP traffic via AppSec. 2. Boundary markers: Absent in the skill's log-handling instructions. 3. Capability inventory: Agent uses cscli with sudo to block IPs and manage firewall decisions. 4. Sanitization: Relies on internal parsers without additional agent-side validation of log content.

crowdsec