product-changelog

Pass

Audited by Gen Agent Trust Hub on Mar 27, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill includes instructions to download and execute a shell script from https://cli.inference.sh to install the infsh CLI. This script is a vendor-provided resource necessary for the skill's primary functionality.
  • [REMOTE_CODE_EXECUTION]: The documentation references npx skills add commands which fetch and execute packages from a remote registry. These packages are part of the vendor's own ecosystem of tools.
  • [COMMAND_EXECUTION]: The skill uses the infsh command to run various cloud-based applications (such as falai/flux-dev-lora and infsh/agent-browser) to generate images and screenshots. The tool access is restricted by the frontmatter configuration.
  • [EXTERNAL_DOWNLOADS]: The skill references https://dist.inference.sh for manual binary installation and the verification of SHA-256 checksums, which are hosted on the vendor's distribution servers.
  • [PROMPT_INJECTION]: The skill features a potential surface for indirect prompt injection by facilitating the retrieval of external web content via the agent-browser tool.
  • Ingestion points: URL parameters passed to the infsh/agent-browser application in the visual generation section.
  • Boundary markers: Absent; there are no defined delimiters to isolate or ignore instructions embedded in the external web pages being browsed.
  • Capability inventory: Execution of the infsh CLI for shell-based tasks, remote package installation via npx, and network-dependent image generation.
  • Sanitization: Not identified; the skill does not provide mechanisms for validating or sanitizing the content retrieved from external URLs before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 27, 2026, 02:31 PM