Agent Paranoid Android¶
Agent Paranoid Android generates deterministic synthetic test datasets from
CSV structure, safe profile metadata, or a reviewed DatasetSpec.
It is designed for cases where realistic schema, relationships, distributions, and business rules matter, but source rows and raw PII must not appear in the generated output.
Choose Your Starting Point¶
| I have | Start here |
|---|---|
| One CSV file | First CSV Dataset |
| A folder of related CSV files | Related Tables |
A safe profile or DatasetSpec |
Profiles And Specs |
| Business constraints | Add Business Rules |
| An MCP-compatible AI client | Connect An MCP Client |
| A failed command | Troubleshooting |
The Safe Workflow¶
- Profile source structure and bounded aggregate metadata.
- Review the inferred
DatasetSpecand any business rules. - Generate fresh values from an explicit seed.
- Validate schema, relationships, constraints, and business rules.
- Review
generation_manifest.jsonbefore accepting the dataset.
The generated bundle reports:
{
"synthetic": true,
"source_rows_copied": false,
"seed": 12345,
"validation_valid": true
}
These fields are evidence produced by deterministic checks. They are not a replacement for reviewing the inferred specification when a new data domain is introduced.
Five-Minute Check¶
Install the package and run the self-contained smoke test:
python3 -m pip install agent-paranoid-android
test-data-agent doctor
A healthy installation ends with:
quickstart smoke: ok
doctor passed
Continue with First CSV Dataset to create and inspect a complete synthetic output bundle.
Safety Boundaries¶
The project intentionally refuses:
- copying or shuffling source rows;
- raw PII or secret values in profiles and rule literals;
- unrestricted SQL or write operations through Trino tools;
- output paths that overwrite source input;
- unbounded input, output, rule, query, or generation work.
Read Safety Model before connecting the project to production-adjacent data or an AI client.
Project Status¶
The current package version is 0.7.0. DatasetSpec is the generation and
validation contract for the CLI and Python API.
Development is substantially AI-assisted. Human review, automated tests, and the documented security requirements still apply to every change.