The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill instructs the agent to execute curl -fsSL https://cli.inference.sh | sh to install the required software.
Evidence: Found in the Quick Start section of SKILL.md.
Risk: The domain inference.sh is not a trusted source. Downloading and immediately executing code from the internet without integrity checks (like checksums) or manual review can lead to full system compromise if the source is malicious or compromised.
[COMMAND_EXECUTION] (MEDIUM): The skill requests broad permissions for the infsh command via allowed-tools: Bash(infsh *).
Evidence: Defined in the YAML frontmatter of SKILL.md.
Risk: This grants the AI agent the ability to run any subcommand of the infsh CLI. Since the CLI itself is installed via an untrusted script, this expands the attack surface for malicious operations on the local host.
[DATA_EXFILTRATION] (LOW): The instructions include a mandatory infsh login step.
Evidence: SKILL.md (Quick Start).
Risk: CLI login commands typically involve the creation or management of sensitive API tokens or session credentials. Using an unverified third-party CLI increases the risk of credential theft or exposure.
[EXTERNAL_DOWNLOADS] (LOW): The skill's primary functionality is dependent on external binaries and scripts from an untrusted third-party domain.
Evidence: References to inference.sh throughout the markdown.
Risk: The skill relies on infrastructure that is not part of the trusted environment, creating a dependency on the security posture of an external entity.

text-to-speech