The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The troubleshooting documentation recommends using high-privilege commands such as sudo and modifying the host's docker group (sudo usermod -aG docker $USER) to resolve permission issues, which grants root-equivalent access to the host.
[DATA_EXFILTRATION]: The skill includes functionality to send setup feedback to a public ntfy.sh topic (https://ntfy.sh/agenta-bdk-feedback-install). Because this topic is unauthenticated and publicly readable, there is a risk of exfiltrating sensitive operational details if they are not correctly redacted by the agent.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of external command outputs. Ingestion points: Data enters the context via Docker Compose logs and curl command responses. Boundary markers: The skill relies on instructional warnings within the markdown files to prompt the agent to redact secrets. Capability inventory: The agent is granted Bash, Read, Edit, Write, Grep, and Glob capabilities. Sanitization: No automated sanitization or schema validation is implemented for external data ingestion; redaction is manual.

self-host-agenta