stata
Pass
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides detailed instructions and examples for executing system-level shell commands using Stata's
shelland!syntax. This includes file system management, Git version control, and running the Stata binary in batch mode. - Evidence found in
SKILL.mdandreferences/external-tools-integration.md. - [REMOTE_CODE_EXECUTION]: Documents the execution of arbitrary Python and R code through the
python:andrcall:environments. These features allow the agent to run scripts in separate runtime environments integrated with Stata. - Evidence found in
references/external-tools-integration.md. - [EXTERNAL_DOWNLOADS]: Fetches Stata packages and datasets from official community registries (SSC) and established academic GitHub repositories.
- Evidence:
net install rdrobust, from("https://raw.githubusercontent.com/rdpackages/rdrobust/master/stata")inpackages/rdrobust.md. - [DATA_EXFILTRATION]: Documents capabilities to connect to external SQL databases (ODBC/JDBC) and ingest data from remote URLs, which could be leveraged for data access or unauthorized export.
- Evidence found in
references/data-import-export.mdandreferences/external-tools-integration.md. - [PROMPT_INJECTION]: The skill documents an indirect prompt injection surface by providing instructions for ingesting untrusted data from external files and databases without explicit sanitization guidelines.
- Ingestion points:
import delimited,import excel, andodbc loadinreferences/data-import-export.md. - Boundary markers: Absent in documentation examples.
- Capability inventory: Shell command execution, arbitrary code execution (Python/R), and file-write operations documented across multiple reference files.
- Sanitization: No specific filtering or escaping mechanisms are discussed.
Audit Metadata