The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents and sets the disable_safety_filter parameter to true by default in its parameters documentation. This provides an explicit instruction to the underlying model to bypass standard content safety filters and remove operational constraints.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data which is interpolated into shell commands, creating a potential vector for indirect prompt injection. * Ingestion points: The voice_script, video_prompt, and voice_prompt parameters in the belt app run command ingest free-form text from the user. * Boundary markers: The data is formatted as JSON within the shell command, but the instructions lack explicit boundary markers or directives for the agent to ignore instructions embedded within these strings. * Capability inventory: The skill has the capability to execute shell commands via the belt CLI tool and interacts with external network services. * Sanitization: There is no evidence of input validation, escaping, or sanitization performed on the user-provided strings before they are passed to the CLI.- [EXTERNAL_DOWNLOADS]: The skill instructs the installation of the belt-sh/cli via npx and references installation scripts hosted on a remote repository (https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md). While these resources belong to the service provider, they introduce an external dependency that executes code on the user's system.

p-video-avatar