The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute a shell script from a remote URL (https://cli.inference.sh) using the curl ... | sh pattern. This is a critical security risk as it executes unverified code with the same privileges as the agent environment.
[COMMAND_EXECUTION]: The skill relies on executing shell commands via the infsh CLI. While the allowed-tools frontmatter attempts to restrict the agent to infsh sub-commands, the installation instructions recommend using curl and sh, which creates a manifest inconsistency that could lead to unauthorized command execution if the platform's constraints are not strictly enforced.
[EXTERNAL_DOWNLOADS]: The skill suggests installing additional functionality using npx skills add, which involves downloading and executing code from external repositories. It also references numerous external URLs for image and audio assets, which may originate from untrusted sources.
[PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection by interpolating user-provided text directly into the --input JSON payloads of the infsh command. Maliciously crafted prompts could attempt to influence the CLI's behavior or the downstream AI model execution.
Ingestion points: User-supplied 'prompt' fields within the JSON input for infsh app run (found in SKILL.md).
Boundary markers: Commands use JSON structures, which provide some data separation but do not fully prevent injection within the string values.
Capability inventory: Subprocess calls via Bash(infsh *) (found in SKILL.md).
Sanitization: None observed beyond basic JSON encapsulation.

ai-video-generation