The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run curl -fsSL https://cli.inference.sh | sh for installation. This is a dangerous pattern that executes unverified code from a remote server, which could be compromised to deliver malicious payloads.\n- [EXTERNAL_DOWNLOADS]: The setup process involves downloading binaries from dist.inference.sh. While the documentation mentions checksum verification, the security of this process relies entirely on the integrity of the initial unauthenticated script download.\n- [COMMAND_EXECUTION]: The skill requires access to the Bash tool to run the infsh CLI and the installation script. This grants the skill the ability to perform any action supported by the underlying shell and the downloaded binary.\n- [CREDENTIALS_UNSAFE]: The inclusion of infsh login demonstrates that the skill handles user credentials or API tokens for the inference.sh platform, creating a risk of credential exfiltration if the environment is not adequately isolated.\n- [PROMPT_INJECTION]: The skill processes external media via URLs (images, audio, video) without implementing boundary markers or sanitization. This provides a surface for indirect prompt injection, as malicious instructions could be embedded in the metadata or content of the files processed by the CLI. Ingestion points: media URLs in CLI inputs; Boundary markers: absent; Capability inventory: Bash tool used for CLI execution; Sanitization: not implemented.

ai-avatar-video