python-sdk
Warn
Audited by Gen Agent Trust Hub on Jul 22, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The documentation in 'references/tool-builder.md' includes an example where 'eval()' is used to process a 'calculate' tool call. Using 'eval()' on untrusted or agent-generated input is a dangerous pattern that can lead to arbitrary code execution.
- [REMOTE_CODE_EXECUTION]: The skill exposes an 'internal_tools' API that allows the agent to enable 'code_execution(True)'. This grants the agent the ability to run arbitrary code within its environment, which is a high-privilege capability.
- [DATA_EXFILTRATION]: The SDK provides multiple mechanisms for data egress, including 'client.upload_file' for local file uploads and 'webhook_tool' for sending data to external HTTP endpoints (e.g., Slack). While these are intended features, they form a significant attack surface for data exfiltration.
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the 'inferencesh' package via pip. The package name aligns with the author's namespace ('infsh-skills'). It also suggests installing other skills using 'npx', which are external dependencies.
- [PROMPT_INJECTION]: The skill is designed to process external and potentially untrusted data from files, search tool results (e.g., via Tavily), and webhook responses. The lack of explicit boundary markers or sanitization in the provided examples makes the system vulnerable to indirect prompt injection, especially given its powerful tool capabilities.
Audit Metadata