The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions for the infsh CLI recommend piping a remote script directly into the shell (curl -fsSL https://cli.inference.sh | sh). This execution of unverified remote code is a significant security risk as the script's contents could be modified to perform malicious actions on the host system.
[DATA_EXFILTRATION]: The CLI documentation highlights a 'Local File Uploads' feature (documented in references/running-apps.md) where the tool automatically reads and uploads any local file path provided in the input JSON to the inference.sh cloud service. This capability creates a data exfiltration surface where an agent, processing untrusted user prompts, might be tricked into reading and uploading sensitive system files (e.g., ~/.ssh/id_rsa, .env, or .aws/credentials) without explicit user consent.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute infsh commands. The documentation suggests commands for system-wide configuration, such as writing shell completions to /etc/bash_completion.d/ (documented in references/cli-reference.md), which typically requires root privileges and could be leveraged for persistence.

agent-tools