The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to execute a remote script directly in the shell using a pipe-to-shell command.
Evidence: curl -fsSL https://cli.inference.sh | sh found in the Quick Start section of SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of a CLI binary and references additional external skill repositories.
Evidence: Mentions downloading from dist.inference.sh and adding related skills from inference-sh/skills using npx in SKILL.md.
[COMMAND_EXECUTION]: The skill requires permission to execute bash commands via the infsh tool to perform its primary functions.
Evidence: allowed-tools: Bash(infsh *) defined in the YAML frontmatter of SKILL.md.
[PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection where untrusted data is interpolated into command arguments.
Ingestion points: The prompt field in various infsh app run commands in SKILL.md allows user-controlled input to be passed to the tool.
Boundary markers: No delimiters or "ignore embedded instructions" warnings are used when processing the prompt input.
Capability inventory: The Bash(infsh *) capability allows the agent to execute a wide variety of inference applications based on the provided prompt.
Sanitization: No sanitization, escaping, or validation logic is defined for the prompt content before it is processed by the CLI.

image-to-video