The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation in SKILL.md and references/agentfs.md recommends installing the AgentFS sandbox by piping a remote script directly to bash: curl -fsSL https://agentfs.ai/install | bash. This allows the remote server to execute arbitrary code on the user's machine without prior verification.
[EXTERNAL_DOWNLOADS]: The skill references downloading software and installation scripts from agentfs.ai and GitHub repositories under the tursodatabase organization. While the documentation provides a safer installation method with checksum verification as an alternative, the primary recommendation remains the high-risk one-liner.
[COMMAND_EXECUTION]: The scripts/swival-proxy bash script manages the litellm proxy process using commands like nohup, ps, and kill to control execution and monitor process IDs.
[PROMPT_INJECTION]: As a coding agent that processes project files and tool outputs, the skill is vulnerable to indirect prompt injection. Although it includes a mechanism to tag outputs as untrusted, the agent's extensive capabilities (filesystem access, command execution) mean that malicious instructions embedded in a project's files could potentially lead to unauthorized actions.

swival