database-review

Pass

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to use local CLI tools such as pg_ctl, psql, and prisma to manage databases and apply migrations. It also suggests using squawk-cli via npm.\n- [EXTERNAL_DOWNLOADS]: Mentions using the squawk-cli npm package from https://squawkhq.com. This is a well-known linting tool for database migrations.\n- [INDIRECT_PROMPT_INJECTION]: The skill takes untrusted migration files or PR diffs as input and executes the contained SQL on a database. While the skill mandates using a 'scratch' database on 'localhost' to mitigate risk, malicious SQL could still attempt to exploit the database engine or the agent's environment if the database is not sufficiently isolated.\n
  • Ingestion points: SKILL.md (Review and Write modes ingest PR diffs, branches, and migration files).\n
  • Boundary markers: Absent. No explicit delimiters are mentioned for wrapping untrusted code before processing.\n
  • Capability inventory: psql, pg_ctl, prisma, and npm (via squawk-cli).\n
  • Sanitization: The skill relies on the agent's review and host-validation logic ('confirm its host before applying') rather than programmatic sanitization of the SQL content.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 16, 2026, 02:34 AM