Skip to content

DeepTeam is a framework to red team LLMs and AI agents.

feat(cli): add scan command

tanayvaswani07a471e6f662
diff --git a/deepteam/cli/main.py b/deepteam/cli/main.py
index ec7ddaef..3cb2f754 100644
--- a/deepteam/cli/main.py
+++ b/deepteam/cli/main.py
@@ -396,5 +396,111 @@ def run(
return result
+def scan(
+ path: str = typer.Argument(".", help="File or directory to scan."),
+ diff: Optional[str] = typer.Option(
+ None,
+ "--diff",
+ help="Scan only files changed between two git refs, e.g. 'main..HEAD'.",
+ ),
+ config_file: Optional[str] = typer.Option(
+ None,
+ "-c",
+ "--config",
+ help="Config YAML path (otherwise auto-discovered at the scan root).",
+ ),
+ output_format: str = typer.Option(
+ "markdown",
+ "-f",
+ "--format",
+ help="Output format: markdown | sarif | json.",
+ ),
+ min_severity: Optional[str] = typer.Option(
+ None,
+ "--min-severity",
+ help="Only report findings at or above: low | medium | high | critical.",
+ ),
+ model: Optional[str] = typer.Option(
+ None,
+ "-m",
+ "--model",
+ help="Evaluation model (e.g. gpt-5.5). Overrides the config.",
+ ),
+ output: Optional[str] = typer.Option(
+ None,
+ "-o",
+ "--output",
+ help="Write the report to a file instead of stdout.",
+ ),
+ fail_on_findings: bool = typer.Option(
+ True,
+ "--fail-on-findings/--no-fail-on-findings",
+ help="Exit with code 1 if any findings remain (useful in CI).",
+ ),
+):
+ """Scan source code for AI-security vulnerabilities."""
+ from deepteam.code_scanner import (
+ CodeScanner,
+ collect_changed_files,
+ collect_files,
+ filter_by_severity,
+ load_config,
+ to_json,
+ to_markdown,
+ to_sarif,
+ )
+
+ formatters = {
+ "markdown": to_markdown,
+ "sarif": to_sarif,
+ "json": to_json,
+ }
+ if output_format not in formatters:
+ raise typer.BadParameter(
+ f"format must be one of {', '.join(formatters)}"
+ )
+
+ config.apply_env()
+
+ scan_root = path if os.path.isdir(path) else os.path.dirname(path) or "."
+ cfg = load_config(path=config_file, directory=scan_root)
+
+ eval_model = load_model(model or cfg.model)
+
+ if diff:
+ base, _, head = diff.partition("..")
+ chunks = collect_changed_files(
+ path,
+ base=base,
+ head=head or None,
+ include=cfg.include,
+ exclude=cfg.exclude,
+ )
+ else:
+ chunks = collect_files(path, include=cfg.include, exclude=cfg.exclude)
+
+ typer.echo(f"Scanning {len(chunks)} code chunk(s)...", err=True)
+
+ scanner = CodeScanner(
+ model=eval_model,
+ vulnerabilities=cfg.vulnerabilities,
+ instruction=cfg.instruction,
+ )
+ findings = scanner.scan(chunks)
+ findings = filter_by_severity(findings, min_severity or cfg.min_severity)
+
+ rendered = formatters[output_format](findings)
+ if output:
+ with open(output, "w") as f:
+ f.write(rendered)
+ typer.echo(f"Wrote {len(findings)} finding(s) to {output}", err=True)
+ else:
+ typer.echo(rendered)
+
+ if fail_on_findings and findings:
+ raise typer.Exit(code=1)
+
+
if __name__ == "__main__":
app()