Static analysis flagged many command, network, environment, and weak-crypto patterns, but manual review found most are Markdown examples or ORM syntax. No prompt injection, data exfiltration, hidden network call, or malicious intent was found. The remaining risk is operational because the skill can guide agents to inspect database metadata and run migration commands that may alter schema state.
Database Migration Commands Require Human Approval
TRUE_POSITIVE operational risk. The Alembic cookbook documents revision, upgrade, downgrade, add-column, and drop-column workflows. These are legitimate for schema alignment, but executing them without explicit approval can change or remove database schema state.
The commands and migration examples are clearly present and can affect schema state. Confidence is below very high because they are documentation examples, not automatically executed code.
Live Database Introspection May Expose Schema Metadata
NEEDS_REVIEW operational risk. The skill allows querying information_schema if accessible and shows a Prisma datasource using DATABASE_URL. This is normal database tooling, but users should avoid exposing credentials or schema metadata outside the project context.
The evidence shows possible database access and a credential-bearing environment variable reference. I did not find network exfiltration or instructions to reveal the connection string.
External Command Detections Are Mostly Markdown Examples
FALSE_POSITIVE for code execution. The static analyzer matched fenced Markdown snippets and inline ORM syntax across the skill. The reviewed files do not contain executable scripts, hooks, or dynamic shell invocation.
The cited lines are documentation blocks rather than runnable skill code. Some commands are still operationally sensitive if a user chooses to run them.
Weak Cryptography Detections Are Documentation False Positives
FALSE_POSITIVE for weak cryptography. The cited locations are skill metadata, headings, Markdown tables, migration placeholders, and JSON field names. No hashing, encryption, signing, or random generation logic was found there.
The reviewed lines contain documentation text and examples, not cryptographic APIs. The static labels appear to come from broad keyword matching.