pixeltable
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill focuses on providing instructions for the Pixeltable database CLI. It prioritizes safety by requiring explicit user approval for destructive operations like drop or recompute, and for remote Cloud operations.
- [SAFE]: The included scripts/preflight.py utility is a read-only script designed to verify file integrity. It uses secure POSIX primitives to avoid symlink traversal and maintains a list of blocked sensitive directories and file patterns.
- [SAFE]: Instructions for credential management are well-defined, advising the agent to use environment variables or secret management tools and strictly forbidding the printing of secret values in output.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect injection as it processes database schemas and computed column data. Ingestion points: Data enters the agent context via pxt ls, pxt describe, pxt columns, and pxt rows commands in the shell. Boundary markers: Instructions require confirming the target repository and obtaining separate user approval for mutations. Capability inventory: The skill uses Bash to execute the pxt CLI, which can perform local file writes (pxt init), database mutations (drop), and remote Cloud operations. Sanitization: The skill advises against printing secrets and recommends using --dry-run to inspect impact before applying changes.
Audit Metadata