iris-security-sdk 0.2.16__tar.gz → 0.2.18__tar.gz

{iris_security_sdk-0.2.16 → iris_security_sdk-0.2.18}/PKG-INFO

@@ -1,11 +1,13 @@
 Metadata-Version: 2.4
 Name: iris-security-sdk
-Version: 0.2.16
+Version: 0.2.18
 Summary: IRIS — AI Agent Governance SDK. Govern AI agents locally.
 Author-email: Gilbert Martin <gilbert@iris-security.io>
 License: Apache-2.0
-Project-URL: Homepage, https://gimartinb.github.io/iris-sdk/
-Project-URL: Repository, https://github.com/gimartinb/iris-sdk
+Project-URL: Homepage, https://iris-security.github.io/iris-sdk/
+Project-URL: Repository, https://github.com/IRIS-Security/iris-sdk
+Project-URL: Documentation, https://iris-security.github.io/iris-sdk/getting-started.html
+Project-URL: Issues, https://github.com/IRIS-Security/iris-sdk/issues
 Keywords: ai-governance,policy-as-code,cedar,compliance,llm,ai-agents,langchain,anthropic,openai,colorado-ai-act,eu-ai-act,hipaa,fedramp,mcp,claude,iris,guardrails,responsible-ai
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Developers
@@ -17,7 +19,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
-Requires-Dist: iris-security-core>=0.1.16
+Requires-Dist: iris-security-core>=0.1.18
 Requires-Dist: pyyaml>=6.0
 Requires-Dist: pydantic>=2.0
 Requires-Dist: click>=8.1
@@ -62,7 +64,7 @@ IRIS — AI Agent Governance SDK. Govern AI agents locally.
 Discover, register, and govern AI agents with Cedar policy evaluation,
 compliance bundles, and local-first governance.
 
-Part of the [IRIS SDK](https://github.com/gimartinb/iris-sdk).
+Part of the [IRIS SDK](https://github.com/IRIS-Security/iris-sdk).
 
 ## Install
 

{iris_security_sdk-0.2.16 → iris_security_sdk-0.2.18}/README.md

@@ -5,7 +5,7 @@ IRIS — AI Agent Governance SDK. Govern AI agents locally.
 Discover, register, and govern AI agents with Cedar policy evaluation,
 compliance bundles, and local-first governance.
 
-Part of the [IRIS SDK](https://github.com/gimartinb/iris-sdk).
+Part of the [IRIS SDK](https://github.com/IRIS-Security/iris-sdk).
 
 ## Install
 

{iris_security_sdk-0.2.16 → iris_security_sdk-0.2.18}/iris/__init__.py

@@ -47,9 +47,9 @@ from iris_core.evidence.vault import EvidenceVault
 from iris_core.cost.tracker import CostSummary, CostTracker, CostEntry
 from iris_core.cost.pricing import PricingRegistry
 
-__version__ = "0.2.16"
+__version__ = "0.2.18"
 
-AARM_CONFORMANCE = "Core"  # R1-R6 satisfied
+AARM_ALIGNMENT = "designed-toward"  # AARM R1–R9 partial; not Core conformant
 AARM_VERSION = "1.0"
 AIUC1_ALIGNMENT = "Q1-2026"
 __all__ = [
@@ -78,7 +78,7 @@ __all__ = [
     "CostSummary",
     "CostEntry",
     "PricingRegistry",
-    "AARM_CONFORMANCE",
+    "AARM_ALIGNMENT",
     "AARM_VERSION",
     "AIUC1_ALIGNMENT",
     "__version__",

{iris_security_sdk-0.2.16 → iris_security_sdk-0.2.18}/iris/scan.py

@@ -57,27 +57,43 @@ _MODEL_PROVIDER_PREFIXES: list[tuple[re.Pattern[str], str]] = [
 ]
 
 _DATA_CATEGORY_PATTERNS = {
+    # Note: regulation/framework names (hipaa, pci, gdpr, ...) are deliberately
+    # excluded here. A compliance tool's own source legitimately talks *about*
+    # those frameworks (rule engines, questionnaire copy, requirement labels)
+    # without ever handling the regulated data itself, so a bare framework
+    # name is not evidence of a data category — only concrete data-field
+    # vocabulary is. Likewise "passport" alone is excluded: this codebase's
+    # own AgentPassport/passport.yaml artifact naming collides with the word
+    # far more often than it means travel-document PII.
     "pii": re.compile(
-        r"\b(ssn|social_security|date_of_birth|dob|passport|email_address|phone_number|"
-        r"first_name|last_name|address_line|national_id)\b",
+        r"\b(ssn|social_security|date_of_birth|dob|passport_number|passport_no|"
+        r"email_address|phone_number|first_name|last_name|address_line|national_id)\b",
         re.I,
     ),
     "phi": re.compile(
-        r"\b(patient_id|medical_record|diagnosis|icd_?10|hipaa|health_record|"
+        r"\b(patient_id|medical_record|diagnosis|icd_?10|health_record|"
         r"prescription|mrn|protected_health)\b",
         re.I,
     ),
     "financial": re.compile(
-        r"\b(credit_card|card_number|bank_account|routing_number|iban|pci|"
+        r"\b(credit_card|card_number|bank_account|routing_number|iban|"
         r"account_balance|transaction_amount)\b",
         re.I,
     ),
     "biometric": re.compile(
-        r"\b(fingerprint|face_id|retina_scan|voice_print|biometric)\b",
+        r"\b(fingerprint|face_id|retina_scan|voice_print)\b",
         re.I,
     ),
 }
 
+# Adjacency required (after removing whitespace/quote noise) for a data-field
+# term to count as source code actually *handling* that field, rather than
+# just naming it in a docstring, questionnaire string, or keyword list — an
+# assignment, a type annotation, or a dict/JSON key all qualify; a bare
+# occurrence inside a comma-separated keyword list or prose sentence does
+# not.
+_CODE_ASSIGNMENT_CONTEXT = re.compile(r"\A\s{0,3}[\'\"]?\s{0,3}[:=]")
+
 _REQUIREMENTS_FILES = (
     "requirements.txt",
     "requirements-dev.txt",
@@ -100,6 +116,16 @@ def _iter_source_files(root: Path) -> list[Path]:
     return files
 
 
+def _is_test_file(path: Path) -> bool:
+    """Test suites routinely embed sample field names/fixtures (fake ssns,
+    demo passport.yaml content, etc.) to exercise detection logic itself —
+    that sample data isn't evidence the workload actually handles it."""
+    if any(part in {"tests", "test"} for part in path.parts):
+        return True
+    stem = path.stem
+    return stem.startswith("test_") or stem.endswith("_test")
+
+
 def _scan_imports(text: str) -> tuple[set[str], set[str]]:
     providers: set[str] = set()
     frameworks: set[str] = set()
@@ -132,6 +158,20 @@ def _scan_data_categories(text: str) -> set[str]:
     return categories
 
 
+def _scan_data_categories_in_source(text: str) -> set[str]:
+    """Like `_scan_data_categories`, but for source files: only count a term
+    that sits where a real field would (assignment, annotation, or dict/JSON
+    key), not one that merely appears in prose, a docstring, or a keyword
+    list describing the vocabulary."""
+    categories: set[str] = set()
+    for category, pattern in _DATA_CATEGORY_PATTERNS.items():
+        for match in pattern.finditer(text):
+            if _CODE_ASSIGNMENT_CONTEXT.match(text[match.end() : match.end() + 6]):
+                categories.add(category)
+                break
+    return categories
+
+
 def infer_provider_from_model(model: str) -> str | None:
     """Infer provider slug from a model identifier (shared by observability adapters)."""
     if not model:
@@ -196,7 +236,8 @@ def detect_workload(path: str = ".") -> dict[str, Any]:
         providers |= p
         frameworks |= f
         models |= _scan_models(text)
-        data_categories |= _scan_data_categories(text)
+        if not _is_test_file(file_path):
+            data_categories |= _scan_data_categories_in_source(text)
 
     agent_hints = sum(1 for _ in root.rglob("passport.yaml"))
     autonomy = "assistive"

{iris_security_sdk-0.2.16 → iris_security_sdk-0.2.18}/iris_security_sdk.egg-info/PKG-INFO

@@ -1,11 +1,13 @@
 Metadata-Version: 2.4
 Name: iris-security-sdk
-Version: 0.2.16
+Version: 0.2.18
 Summary: IRIS — AI Agent Governance SDK. Govern AI agents locally.
 Author-email: Gilbert Martin <gilbert@iris-security.io>
 License: Apache-2.0
-Project-URL: Homepage, https://gimartinb.github.io/iris-sdk/
-Project-URL: Repository, https://github.com/gimartinb/iris-sdk
+Project-URL: Homepage, https://iris-security.github.io/iris-sdk/
+Project-URL: Repository, https://github.com/IRIS-Security/iris-sdk
+Project-URL: Documentation, https://iris-security.github.io/iris-sdk/getting-started.html
+Project-URL: Issues, https://github.com/IRIS-Security/iris-sdk/issues
 Keywords: ai-governance,policy-as-code,cedar,compliance,llm,ai-agents,langchain,anthropic,openai,colorado-ai-act,eu-ai-act,hipaa,fedramp,mcp,claude,iris,guardrails,responsible-ai
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Developers
@@ -17,7 +19,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
-Requires-Dist: iris-security-core>=0.1.16
+Requires-Dist: iris-security-core>=0.1.18
 Requires-Dist: pyyaml>=6.0
 Requires-Dist: pydantic>=2.0
 Requires-Dist: click>=8.1
@@ -62,7 +64,7 @@ IRIS — AI Agent Governance SDK. Govern AI agents locally.
 Discover, register, and govern AI agents with Cedar policy evaluation,
 compliance bundles, and local-first governance.
 
-Part of the [IRIS SDK](https://github.com/gimartinb/iris-sdk).
+Part of the [IRIS SDK](https://github.com/IRIS-Security/iris-sdk).
 
 ## Install
 

{iris_security_sdk-0.2.16 → iris_security_sdk-0.2.18}/iris_security_sdk.egg-info/requires.txt

@@ -1,4 +1,4 @@
-iris-security-core>=0.1.16
+iris-security-core>=0.1.18
 pyyaml>=6.0
 pydantic>=2.0
 click>=8.1

{iris_security_sdk-0.2.16 → iris_security_sdk-0.2.18}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "iris-security-sdk"
-version = "0.2.16"
+version = "0.2.18"
 description = "IRIS — AI Agent Governance SDK. Govern AI agents locally."
 readme = "README.md"
 authors = [{ name = "Gilbert Martin", email = "gilbert@iris-security.io" }]
@@ -27,7 +27,7 @@ classifiers = [
   "Programming Language :: Python :: 3.12",
 ]
 dependencies = [
-  "iris-security-core>=0.1.16",
+  "iris-security-core>=0.1.18",
   "pyyaml>=6.0",
   "pydantic>=2.0",
   "click>=8.1",
@@ -63,8 +63,10 @@ dev = [
 ]
 
 [project.urls]
-Homepage = "https://gimartinb.github.io/iris-sdk/"
-Repository = "https://github.com/gimartinb/iris-sdk"
+Homepage = "https://iris-security.github.io/iris-sdk/"
+Repository = "https://github.com/IRIS-Security/iris-sdk"
+Documentation = "https://iris-security.github.io/iris-sdk/getting-started.html"
+Issues = "https://github.com/IRIS-Security/iris-sdk/issues"
 
 [tool.setuptools.packages.find]
 where = ["."]

iris_security_sdk-0.2.18/tests/test_scan.py

@@ -0,0 +1,125 @@
+# Copyright 2024-2025 Gilbert Martin / IRIS Security, Inc.
+
+from __future__ import annotations
+
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+from iris.scan import detect_workload
+
+
+@pytest.fixture
+def sample_project(tmp_path: Path) -> Path:
+    (tmp_path / "requirements.txt").write_text(
+        "openai>=1.0\nlangchain>=0.2\n",
+        encoding="utf-8",
+    )
+    app = tmp_path / "app.py"
+    app.write_text(
+        """
+import openai
+from langchain_openai import ChatOpenAI
+
+model = ChatOpenAI(model="gpt-4o")
+client = openai.OpenAI()
+
+class UserSchema:
+    ssn = "social_security"
+    email_address = "test@example.com"
+""",
+        encoding="utf-8",
+    )
+    return tmp_path
+
+
+def test_detect_workload_finds_openai_and_langchain(sample_project: Path):
+    profile = detect_workload(str(sample_project))
+    assert "openai" in profile["providers"]
+    assert "langchain" in profile["frameworks"]
+    assert any("gpt" in m for m in profile["models"])
+    assert "pii" in profile["data_categories"]
+
+
+def test_detect_workload_offline_no_network(sample_project: Path):
+    def fail_socket(*args, **kwargs):
+        raise AssertionError("Network call attempted during offline scan")
+
+    with patch("socket.socket", side_effect=fail_socket):
+        profile = detect_workload(str(sample_project))
+    assert profile["source"] == "sdk_scan"
+
+
+# Regression coverage for the dogfood false positive: detect_workload() used
+# to flag data_categories=[biometric, financial, phi, pii] when scanning a
+# governance/compliance tool's own source, because that source legitimately
+# talks *about* those categories (DataClassification enum members, framework
+# names like HIPAA/PCI, questionnaire copy, AgentPassport/passport.yaml
+# vocabulary) without ever handling real data of that kind. apps/iris-cli/ in
+# this repo is exactly that kind of codebase, so scanning it is the natural
+# regression fixture.
+_REPO_ROOT = Path(__file__).resolve().parents[3]
+_IRIS_CLI_ROOT = _REPO_ROOT / "apps" / "iris-cli"
+
+
+@pytest.mark.skipif(not _IRIS_CLI_ROOT.is_dir(), reason="apps/iris-cli not present in this checkout")
+def test_detect_workload_does_not_flag_own_compliance_vocabulary():
+    profile = detect_workload(str(_IRIS_CLI_ROOT))
+    assert profile["data_categories"] == []
+
+
+@pytest.mark.skipif(not _IRIS_CLI_ROOT.is_dir(), reason="apps/iris-cli not present in this checkout")
+def test_detect_workload_does_not_flag_iris_python_own_source():
+    iris_python_root = Path(__file__).resolve().parents[1]
+    profile = detect_workload(str(iris_python_root))
+    assert profile["data_categories"] == []
+
+
+def test_detect_workload_ignores_vocabulary_in_keyword_lists_and_prose(tmp_path: Path):
+    (tmp_path / "linter.py").write_text(
+        '''
+"""Detects high-risk domains in someone else's code."""
+
+high_risk_keywords = ["loan", "credit", "insurance", "medical", "diagnosis"]
+
+QUESTIONS = [
+    "Yes -- biometric or sensitive data",
+    "HIPAA (healthcare)",
+    "PCI DSS (payment cards)",
+]
+
+def load_passport(agent):
+    passport = (agent / "passport.yaml").read_text()
+    return passport
+''',
+        encoding="utf-8",
+    )
+    profile = detect_workload(str(tmp_path))
+    assert profile["data_categories"] == []
+
+
+def test_detect_workload_still_flags_real_field_assignment(tmp_path: Path):
+    (tmp_path / "models.py").write_text(
+        '''
+class Patient:
+    patient_id: str
+    diagnosis = ""
+
+record = {"credit_card": "4111111111111111", "fingerprint": template}
+''',
+        encoding="utf-8",
+    )
+    profile = detect_workload(str(tmp_path))
+    assert set(profile["data_categories"]) == {"phi", "financial", "biometric"}
+
+
+def test_detect_workload_skips_data_category_matches_in_test_files(tmp_path: Path):
+    tests_dir = tmp_path / "tests"
+    tests_dir.mkdir()
+    (tests_dir / "test_fixture.py").write_text(
+        'ssn = "123-45-6789"\ncredit_card = "4111111111111111"\n',
+        encoding="utf-8",
+    )
+    profile = detect_workload(str(tmp_path))
+    assert profile["data_categories"] == []

iris_security_sdk-0.2.16/tests/test_scan.py

@@ -1,51 +0,0 @@
-# Copyright 2024-2025 Gilbert Martin / IRIS Security, Inc.
-
-from __future__ import annotations
-
-from pathlib import Path
-from unittest.mock import patch
-
-import pytest
-
-from iris.scan import detect_workload
-
-
-@pytest.fixture
-def sample_project(tmp_path: Path) -> Path:
-    (tmp_path / "requirements.txt").write_text(
-        "openai>=1.0\nlangchain>=0.2\n",
-        encoding="utf-8",
-    )
-    app = tmp_path / "app.py"
-    app.write_text(
-        """
-import openai
-from langchain_openai import ChatOpenAI
-
-model = ChatOpenAI(model="gpt-4o")
-client = openai.OpenAI()
-
-class UserSchema:
-    ssn = "social_security"
-    email_address = "test@example.com"
-""",
-        encoding="utf-8",
-    )
-    return tmp_path
-
-
-def test_detect_workload_finds_openai_and_langchain(sample_project: Path):
-    profile = detect_workload(str(sample_project))
-    assert "openai" in profile["providers"]
-    assert "langchain" in profile["frameworks"]
-    assert any("gpt" in m for m in profile["models"])
-    assert "pii" in profile["data_categories"]
-
-
-def test_detect_workload_offline_no_network(sample_project: Path):
-    def fail_socket(*args, **kwargs):
-        raise AssertionError("Network call attempted during offline scan")
-
-    with patch("socket.socket", side_effect=fail_socket):
-        profile = detect_workload(str(sample_project))
-    assert profile["source"] == "sdk_scan"