fallow
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONPERSISTENCECOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill recommends installing the
fallowutility through standard package managers likenpmandcargo. These downloads target the official distribution channels for the tool. - [DATA_EXFILTRATION]: The skill communicates with the
fallow.clouddomain to fetch production runtime coverage data. This is a documented feature for merging production execution metrics into the analysis reports. - [INDIRECT_PROMPT_INJECTION]: As the skill ingests and analyzes codebase data and configuration files, it identifies this as a potential attack surface. It includes explicit agent instructions to treat project configuration as untrusted and ignore instructions found within remote config content.
- [DYNAMIC_EXECUTION]: The MCP server provides a
code_executetool for composing analysis snippets. This tool operates within a restricted sandbox that removes access to the filesystem, network, and unsafe JavaScript constructors likeevalandFunction. - [PERSISTENCE]: The skill includes commands for installing Git pre-commit hooks and agent-specific hooks (e.g., for Claude Code). These tools are used to maintain code quality standards across the development lifecycle.
- [COMMAND_EXECUTION]: The skill documentation describes the execution of various
fallowCLI commands for auditing, refactoring, and quality gating. These operations are essential to the primary purpose of the skill.
Audit Metadata