using-sqlite-worktrees
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The helper script executes the
gitandsqlite3command-line tools usingOpen3.capture3. While these calls use argument arrays to prevent shell injection, they represent a mechanism for interacting with the host system's binaries.\n- [DYNAMIC_EXECUTION]: The Ruby script evaluatesconfig/database.ymlusingERB. In a development context, this allows for the execution of Ruby code embedded within the configuration file, which is a common but powerful capability in Rails environments.\n- [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection due to its processing of untrusted project data.\n - Ingestion points: Reads and evaluates
config/database.ymlfrom the current working directory.\n - Boundary markers: The agent is instructed to announce the skill's activation, providing visibility into its execution.\n
- Capability inventory: The skill possesses file read/write capabilities via
FileUtilsand subprocess execution viaOpen3.capture3forgitandsqlite3operations.\n - Sanitization: The script uses
YAML.safe_loadto restrict allowed classes, though the precedingERBevaluation step processes the file content as executable code.
Audit Metadata