Skip to content

Safety

Safety scanning engine and test suites.

Scanner

agentprobe.safety.scanner

Safety scanner engine: runs safety test suites against agents.

Provides the SafetyScanner orchestrator and base SafetySuite ABC for implementing specific safety test suites (prompt injection, data leakage, etc.).

SafetySuiteResult

Bases: BaseModel

Result from a single safety suite execution.

Attributes:

Name Type Description
suite_name str

Name of the safety suite.
total_tests int

Number of tests in the suite.
passed int

Number of tests that passed (no safety violation).
failed int

Number of tests that detected a safety issue.
results tuple[dict[str, Any], ...]

Detailed per-test results.
Source code in src/agentprobe/safety/scanner.py 
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
class SafetySuiteResult(BaseModel):
    """Result from a single safety suite execution.

    Attributes:
        suite_name: Name of the safety suite.
        total_tests: Number of tests in the suite.
        passed: Number of tests that passed (no safety violation).
        failed: Number of tests that detected a safety issue.
        results: Detailed per-test results.
    """

    model_config = ConfigDict(strict=True, frozen=True, extra="forbid")

    suite_name: str
    total_tests: int = Field(default=0, ge=0)
    passed: int = Field(default=0, ge=0)
    failed: int = Field(default=0, ge=0)
    results: tuple[dict[str, Any], ...] = ()

SafetyScanResult

Bases: BaseModel

Aggregate result from all safety suites.

Attributes:

Name Type Description
total_suites int

Number of suites executed.
total_tests int

Total tests across all suites.
total_passed int

Tests that passed across all suites.
total_failed int

Tests that failed across all suites.
suite_results tuple[SafetySuiteResult, ...]

Per-suite results.
Source code in src/agentprobe/safety/scanner.py 
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
class SafetyScanResult(BaseModel):
    """Aggregate result from all safety suites.

    Attributes:
        total_suites: Number of suites executed.
        total_tests: Total tests across all suites.
        total_passed: Tests that passed across all suites.
        total_failed: Tests that failed across all suites.
        suite_results: Per-suite results.
    """

    model_config = ConfigDict(strict=True, frozen=True, extra="forbid")

    total_suites: int = Field(default=0, ge=0)
    total_tests: int = Field(default=0, ge=0)
    total_passed: int = Field(default=0, ge=0)
    total_failed: int = Field(default=0, ge=0)
    suite_results: tuple[SafetySuiteResult, ...] = ()

SafetySuite

Bases: ABC

Abstract base class for safety test suites.

Each suite implements a specific category of safety testing (e.g. prompt injection, data leakage, bias detection).

Source code in src/agentprobe/safety/scanner.py 
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
class SafetySuite(ABC):
    """Abstract base class for safety test suites.

    Each suite implements a specific category of safety testing
    (e.g. prompt injection, data leakage, bias detection).
    """

    @property
    @abstractmethod
    def name(self) -> str:
        """Return the suite name."""
        ...

    @abstractmethod
    async def run(self, adapter: AdapterProtocol) -> SafetySuiteResult:
        """Execute all tests in this suite against an adapter.

        Args:
            adapter: The agent adapter to test.

        Returns:
            Results from the suite execution.
        """
        ...

name abstractmethod property

Return the suite name.

run(adapter) abstractmethod async

Execute all tests in this suite against an adapter.

Parameters:

Name Type Description Default
adapter AdapterProtocol

The agent adapter to test.

 required

Returns:

Type Description
SafetySuiteResult

Results from the suite execution.
Source code in src/agentprobe/safety/scanner.py 
74
75
76
77
78
79
80
81
82
83
84
@abstractmethod
async def run(self, adapter: AdapterProtocol) -> SafetySuiteResult:
    """Execute all tests in this suite against an adapter.

    Args:
        adapter: The agent adapter to test.

    Returns:
        Results from the suite execution.
    """
    ...

SafetyScanner

Orchestrates safety testing by running configured suites.

Attributes:

Name Type Description
suites

List of safety suite instances to run.
Source code in src/agentprobe/safety/scanner.py 
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
class SafetyScanner:
    """Orchestrates safety testing by running configured suites.

    Attributes:
        suites: List of safety suite instances to run.
    """

    def __init__(self, suites: list[SafetySuite] | None = None) -> None:
        """Initialize the safety scanner.

        Args:
            suites: Safety suites to run. If None, uses an empty list.
        """
        self._suites = suites or []

    @classmethod
    def from_config(cls, suite_names: list[str]) -> SafetyScanner:
        """Create a scanner from a list of suite names.

        Looks up suite classes in the global registry.

        Args:
            suite_names: Names of suites to instantiate.

        Returns:
            A configured SafetyScanner.
        """
        suites: list[SafetySuite] = []
        for name in suite_names:
            suite_class = _suite_registry.get(name)
            if suite_class is None:
                logger.warning("Unknown safety suite: %s", name)
                continue
            suites.append(suite_class())
        return cls(suites=suites)

    async def scan(self, adapter: AdapterProtocol) -> SafetyScanResult:
        """Run all configured safety suites against an adapter.

        Args:
            adapter: The agent adapter to test.

        Returns:
            Aggregate scan results.
        """
        suite_results: list[SafetySuiteResult] = []

        for suite in self._suites:
            try:
                result = await suite.run(adapter)
                suite_results.append(result)
            except Exception:
                logger.exception("Safety suite '%s' failed", suite.name)
                suite_results.append(
                    SafetySuiteResult(
                        suite_name=suite.name,
                        total_tests=0,
                        passed=0,
                        failed=0,
                    )
                )

        total_tests = sum(r.total_tests for r in suite_results)
        total_passed = sum(r.passed for r in suite_results)
        total_failed = sum(r.failed for r in suite_results)

        return SafetyScanResult(
            total_suites=len(suite_results),
            total_tests=total_tests,
            total_passed=total_passed,
            total_failed=total_failed,
            suite_results=tuple(suite_results),
        )

__init__(suites=None)

Initialize the safety scanner.

Parameters:

Name Type Description Default
suites list[SafetySuite] | None

Safety suites to run. If None, uses an empty list.

 None
Source code in src/agentprobe/safety/scanner.py 
117
118
119
120
121
122
123
def __init__(self, suites: list[SafetySuite] | None = None) -> None:
    """Initialize the safety scanner.

    Args:
        suites: Safety suites to run. If None, uses an empty list.
    """
    self._suites = suites or []

from_config(suite_names) classmethod

Create a scanner from a list of suite names.

Looks up suite classes in the global registry.

Parameters:

Name Type Description Default
suite_names list[str]

Names of suites to instantiate.

 required

Returns:

Type Description
SafetyScanner

A configured SafetyScanner.
Source code in src/agentprobe/safety/scanner.py 
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
@classmethod
def from_config(cls, suite_names: list[str]) -> SafetyScanner:
    """Create a scanner from a list of suite names.

    Looks up suite classes in the global registry.

    Args:
        suite_names: Names of suites to instantiate.

    Returns:
        A configured SafetyScanner.
    """
    suites: list[SafetySuite] = []
    for name in suite_names:
        suite_class = _suite_registry.get(name)
        if suite_class is None:
            logger.warning("Unknown safety suite: %s", name)
            continue
        suites.append(suite_class())
    return cls(suites=suites)

scan(adapter) async

Run all configured safety suites against an adapter.

Parameters:

Name Type Description Default
adapter AdapterProtocol

The agent adapter to test.

 required

Returns:

Type Description
SafetyScanResult

Aggregate scan results.
Source code in src/agentprobe/safety/scanner.py 
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
async def scan(self, adapter: AdapterProtocol) -> SafetyScanResult:
    """Run all configured safety suites against an adapter.

    Args:
        adapter: The agent adapter to test.

    Returns:
        Aggregate scan results.
    """
    suite_results: list[SafetySuiteResult] = []

    for suite in self._suites:
        try:
            result = await suite.run(adapter)
            suite_results.append(result)
        except Exception:
            logger.exception("Safety suite '%s' failed", suite.name)
            suite_results.append(
                SafetySuiteResult(
                    suite_name=suite.name,
                    total_tests=0,
                    passed=0,
                    failed=0,
                )
            )

    total_tests = sum(r.total_tests for r in suite_results)
    total_passed = sum(r.passed for r in suite_results)
    total_failed = sum(r.failed for r in suite_results)

    return SafetyScanResult(
        total_suites=len(suite_results),
        total_tests=total_tests,
        total_passed=total_passed,
        total_failed=total_failed,
        suite_results=tuple(suite_results),
    )

register_suite(suite_class)

Register a safety suite class in the global registry.

Parameters:

Name Type Description Default
suite_class type[SafetySuite]

The suite class to register.

 required

Returns:

Type Description
type[SafetySuite]

The same class (for use as a decorator).
Source code in src/agentprobe/safety/scanner.py 
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
def register_suite(suite_class: type[SafetySuite]) -> type[SafetySuite]:
    """Register a safety suite class in the global registry.

    Args:
        suite_class: The suite class to register.

    Returns:
        The same class (for use as a decorator).
    """
    instance = suite_class()
    _suite_registry[instance.name] = suite_class
    return suite_class

get_registered_suites()

Return all registered safety suite classes.

Source code in src/agentprobe/safety/scanner.py 
105
106
107
def get_registered_suites() -> dict[str, type[SafetySuite]]:
    """Return all registered safety suite classes."""
    return dict(_suite_registry)