drizzle-migration-conflict
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes migration files, journal metadata, and configuration files from the user's repository which could contain malicious instructions.
- Ingestion points: Processes
_journal.json,snapshot.json,drizzle.config.*, and SQL migration files (.sql) within the repository workspace. - Boundary markers: The skill includes a 'Safety rules' section and separate operational modes ('Diagnose' vs 'Repair') to ensure the agent remains in a restricted state when handling untrusted data.
- Capability inventory: Access to command-line tools including
git,rg,python3, anddrizzle-kit. It has the ability to modify files in 'Repair' mode. - Sanitization: Structural validation is performed by a Python script that uses safe JSON parsing and regular expressions to inspect files without executing their content.
- [COMMAND_EXECUTION]: The skill facilitates the execution of various system and project-specific commands.
- Evidence: Executes
git status,git ls-files,rg, and the project's owndrizzle-kitcommands. It also runs a provided Python helper scriptscripts/check_drizzle_migrations.pyto perform read-only structural checks.
Audit Metadata