The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions recommend installing the CLI using curl -fsSL https://cli.inference.sh | sh. This is a high-risk pattern that executes remote code directly in the system shell without user inspection.
[EXTERNAL_DOWNLOADS]: The installation process downloads executable binaries from dist.inference.sh. Additionally, the skill suggests adding more features using npx skills add, which involves downloading and executing external packages.
[COMMAND_EXECUTION]: The skill's core functionality relies on executing infsh commands via the Bash tool to interact with remote AI models and automate tasks.
[DATA_EXFILTRATION]: The CLI documentation highlights an 'automatic local file upload' feature. When a local file path is provided in an input field, the CLI automatically uploads the file to the vendor's cloud service. This capability could be exploited to exfiltrate sensitive data if the agent is instructed to process sensitive files.
[PROMPT_INJECTION]: The skill ingests and processes data from various external AI models and search engines, creating a surface for indirect prompt injection where untrusted content could influence the agent's behavior.
Ingestion points: Outputs from infsh app run across multiple AI models (documented in SKILL.md and references/running-apps.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded instructions are present in the examples or prompts.
Capability inventory: Subprocess execution via Bash tool, file reading, and network access.
Sanitization: No evidence of validation or sanitization of external model outputs before processing.

infsh-cli