The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions in SKILL.md, references/authentication.md, and references/cli-reference.md explicitly direct the agent to execute curl -fsSL https://cli.inference.sh | sh. This pattern downloads a script from an external, untrusted URL and pipes it directly into the shell for execution, which is a high-risk vector for system compromise.
[EXTERNAL_DOWNLOADS]: The installation process involves downloading binaries and manifest files from dist.inference.sh. Since this domain and the associated organization are not on the trusted vendors list, these external dependencies cannot be verified for safety.
[COMMAND_EXECUTION]: The skill defines Bash(infsh *) in its allowed-tools. This grants the agent the ability to execute any command within the infsh CLI suite, which includes operations for authentication, file saving, and cloud-based application execution.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to how it processes external data.
Ingestion points: The skill reads untrusted data from local JSON files or inline JSON strings (e.g., infsh app run <app> --input input.json) as shown in references/running-apps.md.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the input data.
Capability inventory: The agent has the capability to execute shell commands and write files to the local system via the infsh CLI.
Sanitization: No evidence of sanitization or validation of the input content is provided before it is passed to the execution tool.

agent-tools