Database Replication Advisor

• Outcome: Review
• Latest audit: May 29, 2026 at 9:38 PM UTC
• Version:
  1.0.1

Overview

This is a coherent database operations guide, but it includes live write and failover commands that could change or disrupt a production database without strong safety gates.

Install only if you want an agent-assisted database operations runbook and will keep execution under human control. Treat the heartbeat, failover, promotion, ALTER SYSTEM, AWS RDS failover, table create/drop, and pg_rewind examples as active operational procedures: use test environments or dedicated test objects, least-privileged credentials, backups, change approval, and a maintenance window before running them.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Description-Behavior Mismatch

Medium

• Confidence: 96% confidence
• Finding: The `assess` workflow is presented as a health check, but it includes an application-level heartbeat that creates a table and performs inserts/updates on the primary database. In an agent setting, a supposedly read-only assessment that mutates production state can violate change-control expectations and may trigger side effects, especially on sensitive or tightly controlled databases.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: The `failover-test` section claims to plan and validate a drill, but it provides commands that actively perform switchover/failover and alter database configuration. If an agent executes these during analysis or planning, it could cause real service disruption, forced role changes, or unintended outage in production.

Intent-Code Divergence

Medium

• Confidence: 95% confidence
• Finding: There is a mismatch between the stated purpose ('plan and validate a failover drill') and the actual content, which performs the failover. This kind of semantic mismatch is dangerous for autonomous or semi-autonomous agents because users may invoke a planning skill expecting analysis only, while the skill contains live operational actions.

Missing User Warnings

Medium

• Confidence: 98% confidence
• Finding: The lag assessment writes to the database by creating `_repl_heartbeat` and updating it, but the skill does not warn that this modifies data. In production environments, undisclosed writes can breach operational policy, create audit noise, interfere with permissions assumptions, and surprise users who requested a passive health check.

Missing User Warnings

High

• Confidence: 97% confidence
• Finding: The failover validation step creates and drops a table on the new primary to test write capability without an explicit warning. Even if transient, this is a destructive schema-changing operation that can fail, lock metadata, trigger replication or auditing side effects, and is especially risky immediately after failover when systems may still be unstable.

VirusTotal

65/65 vendors flagged this skill as clean.