Security audit
Security checks for vulnerabilities and agentic risk
This database skill needs review because its required tests lack isolation safeguards and changes automatically trigger full database exports.
Use a disposable database copy until rollback rehearsals and constraint tests require isolation and reliable cleanup. Make full database exports opt-in, and review their contents and delivery channel: exported database files retain personal data even when snapshots are masked.
Referenced artifact was not completely inspected
md to produce DDL, ERD, seed scripts, and validation; runtime operations follow references/operations.md (intent → SQL → safe execution → natural-language repor
Referenced artifact was not completely inspected
SQL generation → safe execution → natural-language reporting**. Full rules in `references/operations.md` (trigger rules, SQL generation rules, safety gates, re
Referenced artifact was not completely inspected
l data snapshot — rules in the "Post-Change Visualization Delivery" section of `references/operations.md`
Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.
When the user provides files that are **not** CSV/Excel/JSON/JSONL/TSV (e.g. PDF textbooks, PPTX, DOCX, images), never enter the build pipeline directly and never refuse service — follow the triage in `references/intake.md`:
The skill mandates executing the last migration's down script and then reapplying it without requiring an isolated test database, a verified backup, or approval for destructive operations. In this database-management context, a down migration can drop populated tables or columns, and rerunning the up migration generally restores only structure—not deleted data; the migration ledger does not prevent this loss.
Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.
# Non-Data File Intake (intent triage)
When the user provides files that are not CSV/Excel/JSON/JSONL/TSV (PDF, PPTX, DOCX, images, plain text, books, etc.), **never enter the build pipeline directly, and never refuse service**. Follow this triage.
## Triage Decision
Without declared permissions the skill's intent is opaque and cannot be validated.
Automatically delivering the entire database after every change can disclose sensitive or unrelated records beyond the scope of the user's requested operation. The stated snapshot masking does not protect the separately delivered raw database, and shipping audit logs or encoding schema details in online visualization links creates additional disclosure risks depending on their contents and handling.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
## 2. Data Quality Gates
After seeding, automatically run a set of assertions; on failure, report and stop the pipeline:
| Assertion | Default threshold |
|---|---|
The methodology instructs the agent to run negative INSERT tests whenever a database is available, without requiring an isolated test database or rollback. If a constraint is missing or ineffective—the condition these tests are meant to detect—invalid rows can persist; test attempts can also trigger audit entries or other side effects. This is particularly relevant for a skill that performs real database operations, although the provided file does not establish that production access is available.
Every write automatically triggers a full database export and a multi-table HTML snapshot, even when the user requested only a narrow modification. These artifacts can duplicate sensitive or unrelated records into output files or delivery channels with weaker access controls, without a sensitivity check or explicit export consent. Database delivery is legitimate for this skill, but making full exports automatic creates unnecessary exposure; the text does not establish actual transmission to an unauthorized third party.
No suspicious patterns detected.