cap-apps-sql-query

Pass

Audited by Gen Agent Trust Hub on Mar 24, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses AIClient to generate SQL from natural language input which is then executed via SqlClient. This represents a vulnerability surface for indirect prompt injection. 1. Ingestion points: The question input to AIClient.text_to_sql (SKILL.md). 2. Boundary markers: Absent; no delimiters or instructions to ignore embedded commands are present in the documentation. 3. Capability inventory: SqlClient.get executes raw SQL against datasets mapped in manifest.json (SKILL.md). 4. Sanitization: No sanitization, validation, or structural checks of the resulting SQL string are shown before execution.
  • [DATA_EXFILTRATION]: Maliciously crafted questions could potentially influence the generated SQL query to access or expose records from datasets that exceed the user's intended scope or authorization level.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 24, 2026, 04:52 PM