SafeAI-Static-Analyzer 2.5.0__tar.gz → 2.7.0__tar.gz

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: SafeAI-Static-Analyzer
-Version: 2.5.0
+Version: 2.7.0
 Summary: The lightweight, open-source Static AI Capability & Risk Analyzer for CI/CD pipelines.
 Author: IkarusCareer
 License-Expression: Apache-2.0
@@ -23,7 +23,7 @@ Requires-Dist: PyYAML>=6.0.3
 Provides-Extra: dev
 Requires-Dist: pytest>=8.0; extra == "dev"
 Requires-Dist: pytest-cov>=5.0; extra == "dev"
-Requires-Dist: ruff<0.17,>=0.16.1; extra == "dev"
+Requires-Dist: ruff<0.18,>=0.16.1; extra == "dev"
 Requires-Dist: jinja2>=3.1.4; extra == "dev"
 Requires-Dist: jsonschema>=4.22.0; extra == "dev"
 Requires-Dist: typing-extensions>=4.4.0; extra == "dev"
@@ -35,8 +35,9 @@ Dynamic: license-file
 
 [![CI](https://github.com/ikaruscareer/SafeAI/actions/workflows/ci.yml/badge.svg)](https://github.com/ikaruscareer/SafeAI/actions/workflows/ci.yml)
 [![OpenSSF Scorecard](https://api.scorecard.dev/projects/github.com/ikaruscareer/SafeAI/badge)](https://scorecard.dev/viewer/?uri=github.com/ikaruscareer/SafeAI)
-[![Website](https://img.shields.io/badge/web-safeai--analyzer.ikaruscareer.com-0f766e)](https://safeai-analyzer.ikaruscareer.com)
-[![Latest Release](https://img.shields.io/badge/latest-v2.4.0-0f766e)](https://github.com/ikaruscareer/SafeAI/releases/tag/v2.4.0)
+[![Research Hub](https://img.shields.io/badge/research-ikaruslabs.net-0f766e)](https://ikaruslabs.net)
+[![SafeAI](https://img.shields.io/badge/SafeAI-ikaruslabs.net/safeai-0f766e)](https://ikaruslabs.net/safeai)
+[![Latest Release](https://img.shields.io/badge/latest-v2.7.0-0f766e)](https://github.com/ikaruscareer/SafeAI/releases/tag/v2.7.0)
 [![Best Practices](https://bestpractices.dev/projects/14126/badge)](https://www.bestpractices.dev/en/projects/14126)
 
 Enjoying SafeAI? A ⭐ on [GitHub](https://github.com/ikaruscareer/SafeAI) helps more security teams find it.
@@ -55,7 +56,7 @@ Enjoying SafeAI? A ⭐ on [GitHub](https://github.com/ikaruscareer/SafeAI) helps
 
 > SafeAI is not a runtime guardrail and does not certify an agent as safe. It provides source-first, evidence-backed visibility into agent authority and meaningful changes *before deployment*.
 
-> 🌐 [safeai-analyzer.ikaruscareer.com](https://safeai-analyzer.ikaruscareer.com) — project landing page
+> 🌐 [Ikarus AI Labs Research Hub](https://ikaruslabs.net) — SafeAI: [ikaruslabs.net/safeai](https://ikaruslabs.net/safeai) | OpenPulse: [ikaruslabs.net/openpulse](https://ikaruslabs.net/openpulse) | Research: [ikaruslabs.net/research](https://ikaruslabs.net/research)
 
 <img width="1024" height="1024" alt="SafeAI_Agent_Software_Static_Analyzer" src="https://github.com/user-attachments/assets/de40f40b-14b9-4cd6-bc2c-27e81e8253fe" />
 
@@ -608,6 +609,7 @@ contributors who have helped make AI safer:
 | [@chenzeyan54-commits](https://github.com/chenzeyan54-commits) | Terminal output sanitization against ANSI/OSC injection (PR #203) |
 | [@HarshRajSinghania](https://github.com/HarshRajSinghania) | PR comment markdown injection sanitization (PR #204) |
 | [@thadidaniel-ctrl](https://github.com/thadidaniel-ctrl) | Release workflow supply-chain posture tests (PR #206) |
+| [@Golden007-prog](https://github.com/Golden007-prog) | Authority Evidence Benchmark HTML presentation, history limit (PRs #239, #240) |
 
 See [CONTRIBUTING.md](./CONTRIBUTING.md) for how to get involved.
 

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/README.md

@@ -2,8 +2,9 @@
 
 [![CI](https://github.com/ikaruscareer/SafeAI/actions/workflows/ci.yml/badge.svg)](https://github.com/ikaruscareer/SafeAI/actions/workflows/ci.yml)
 [![OpenSSF Scorecard](https://api.scorecard.dev/projects/github.com/ikaruscareer/SafeAI/badge)](https://scorecard.dev/viewer/?uri=github.com/ikaruscareer/SafeAI)
-[![Website](https://img.shields.io/badge/web-safeai--analyzer.ikaruscareer.com-0f766e)](https://safeai-analyzer.ikaruscareer.com)
-[![Latest Release](https://img.shields.io/badge/latest-v2.4.0-0f766e)](https://github.com/ikaruscareer/SafeAI/releases/tag/v2.4.0)
+[![Research Hub](https://img.shields.io/badge/research-ikaruslabs.net-0f766e)](https://ikaruslabs.net)
+[![SafeAI](https://img.shields.io/badge/SafeAI-ikaruslabs.net/safeai-0f766e)](https://ikaruslabs.net/safeai)
+[![Latest Release](https://img.shields.io/badge/latest-v2.7.0-0f766e)](https://github.com/ikaruscareer/SafeAI/releases/tag/v2.7.0)
 [![Best Practices](https://bestpractices.dev/projects/14126/badge)](https://www.bestpractices.dev/en/projects/14126)
 
 Enjoying SafeAI? A ⭐ on [GitHub](https://github.com/ikaruscareer/SafeAI) helps more security teams find it.
@@ -22,7 +23,7 @@ Enjoying SafeAI? A ⭐ on [GitHub](https://github.com/ikaruscareer/SafeAI) helps
 
 > SafeAI is not a runtime guardrail and does not certify an agent as safe. It provides source-first, evidence-backed visibility into agent authority and meaningful changes *before deployment*.
 
-> 🌐 [safeai-analyzer.ikaruscareer.com](https://safeai-analyzer.ikaruscareer.com) — project landing page
+> 🌐 [Ikarus AI Labs Research Hub](https://ikaruslabs.net) — SafeAI: [ikaruslabs.net/safeai](https://ikaruslabs.net/safeai) | OpenPulse: [ikaruslabs.net/openpulse](https://ikaruslabs.net/openpulse) | Research: [ikaruslabs.net/research](https://ikaruslabs.net/research)
 
 <img width="1024" height="1024" alt="SafeAI_Agent_Software_Static_Analyzer" src="https://github.com/user-attachments/assets/de40f40b-14b9-4cd6-bc2c-27e81e8253fe" />
 
@@ -575,6 +576,7 @@ contributors who have helped make AI safer:
 | [@chenzeyan54-commits](https://github.com/chenzeyan54-commits) | Terminal output sanitization against ANSI/OSC injection (PR #203) |
 | [@HarshRajSinghania](https://github.com/HarshRajSinghania) | PR comment markdown injection sanitization (PR #204) |
 | [@thadidaniel-ctrl](https://github.com/thadidaniel-ctrl) | Release workflow supply-chain posture tests (PR #206) |
+| [@Golden007-prog](https://github.com/Golden007-prog) | Authority Evidence Benchmark HTML presentation, history limit (PRs #239, #240) |
 
 See [CONTRIBUTING.md](./CONTRIBUTING.md) for how to get involved.
 

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/SafeAI_Static_Analyzer.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: SafeAI-Static-Analyzer
-Version: 2.5.0
+Version: 2.7.0
 Summary: The lightweight, open-source Static AI Capability & Risk Analyzer for CI/CD pipelines.
 Author: IkarusCareer
 License-Expression: Apache-2.0
@@ -23,7 +23,7 @@ Requires-Dist: PyYAML>=6.0.3
 Provides-Extra: dev
 Requires-Dist: pytest>=8.0; extra == "dev"
 Requires-Dist: pytest-cov>=5.0; extra == "dev"
-Requires-Dist: ruff<0.17,>=0.16.1; extra == "dev"
+Requires-Dist: ruff<0.18,>=0.16.1; extra == "dev"
 Requires-Dist: jinja2>=3.1.4; extra == "dev"
 Requires-Dist: jsonschema>=4.22.0; extra == "dev"
 Requires-Dist: typing-extensions>=4.4.0; extra == "dev"
@@ -35,8 +35,9 @@ Dynamic: license-file
 
 [![CI](https://github.com/ikaruscareer/SafeAI/actions/workflows/ci.yml/badge.svg)](https://github.com/ikaruscareer/SafeAI/actions/workflows/ci.yml)
 [![OpenSSF Scorecard](https://api.scorecard.dev/projects/github.com/ikaruscareer/SafeAI/badge)](https://scorecard.dev/viewer/?uri=github.com/ikaruscareer/SafeAI)
-[![Website](https://img.shields.io/badge/web-safeai--analyzer.ikaruscareer.com-0f766e)](https://safeai-analyzer.ikaruscareer.com)
-[![Latest Release](https://img.shields.io/badge/latest-v2.4.0-0f766e)](https://github.com/ikaruscareer/SafeAI/releases/tag/v2.4.0)
+[![Research Hub](https://img.shields.io/badge/research-ikaruslabs.net-0f766e)](https://ikaruslabs.net)
+[![SafeAI](https://img.shields.io/badge/SafeAI-ikaruslabs.net/safeai-0f766e)](https://ikaruslabs.net/safeai)
+[![Latest Release](https://img.shields.io/badge/latest-v2.7.0-0f766e)](https://github.com/ikaruscareer/SafeAI/releases/tag/v2.7.0)
 [![Best Practices](https://bestpractices.dev/projects/14126/badge)](https://www.bestpractices.dev/en/projects/14126)
 
 Enjoying SafeAI? A ⭐ on [GitHub](https://github.com/ikaruscareer/SafeAI) helps more security teams find it.
@@ -55,7 +56,7 @@ Enjoying SafeAI? A ⭐ on [GitHub](https://github.com/ikaruscareer/SafeAI) helps
 
 > SafeAI is not a runtime guardrail and does not certify an agent as safe. It provides source-first, evidence-backed visibility into agent authority and meaningful changes *before deployment*.
 
-> 🌐 [safeai-analyzer.ikaruscareer.com](https://safeai-analyzer.ikaruscareer.com) — project landing page
+> 🌐 [Ikarus AI Labs Research Hub](https://ikaruslabs.net) — SafeAI: [ikaruslabs.net/safeai](https://ikaruslabs.net/safeai) | OpenPulse: [ikaruslabs.net/openpulse](https://ikaruslabs.net/openpulse) | Research: [ikaruslabs.net/research](https://ikaruslabs.net/research)
 
 <img width="1024" height="1024" alt="SafeAI_Agent_Software_Static_Analyzer" src="https://github.com/user-attachments/assets/de40f40b-14b9-4cd6-bc2c-27e81e8253fe" />
 
@@ -608,6 +609,7 @@ contributors who have helped make AI safer:
 | [@chenzeyan54-commits](https://github.com/chenzeyan54-commits) | Terminal output sanitization against ANSI/OSC injection (PR #203) |
 | [@HarshRajSinghania](https://github.com/HarshRajSinghania) | PR comment markdown injection sanitization (PR #204) |
 | [@thadidaniel-ctrl](https://github.com/thadidaniel-ctrl) | Release workflow supply-chain posture tests (PR #206) |
+| [@Golden007-prog](https://github.com/Golden007-prog) | Authority Evidence Benchmark HTML presentation, history limit (PRs #239, #240) |
 
 See [CONTRIBUTING.md](./CONTRIBUTING.md) for how to get involved.
 

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/SafeAI_Static_Analyzer.egg-info/SOURCES.txt

@@ -22,6 +22,7 @@ safeai/analysis/components.py
 safeai/analysis/dependency_correlation.py
 safeai/analysis/escalation.py
 safeai/analysis/escalation_remediation.py
+safeai/analysis/evidence_metrics.py
 safeai/analysis/iac_correlation.py
 safeai/analysis/import_graph.py
 safeai/analysis/project_graph.py
@@ -60,6 +61,12 @@ safeai/analyzers/tool_def/__init__.py
 safeai/analyzers/tool_def/analyzer.py
 safeai/analyzers/workflow/__init__.py
 safeai/analyzers/workflow/analyzer.py
+safeai/benchmark/__init__.py
+safeai/benchmark/canonical.py
+safeai/benchmark/compare.py
+safeai/benchmark/metrics.py
+safeai/benchmark/report.py
+safeai/benchmark/runner.py
 safeai/cmd/__init__.py
 safeai/cmd/cli.py
 safeai/cmd/manifest_cli.py
@@ -152,6 +159,7 @@ safeai/report/json_report.py
 safeai/report/pr_comment.py
 safeai/report/registry_html.py
 safeai/report/sarif.py
+safeai/report/security_brief.py
 safeai/report/terminal.py
 safeai/rules/__init__.py
 safeai/rules/base_rules.yaml
@@ -167,6 +175,8 @@ tests/test_adapter_dataflow.py
 tests/test_architectural_invariants.py
 tests/test_architecture_diagram.py
 tests/test_assurance.py
+tests/test_authority_benchmark.py
+tests/test_authority_canonical.py
 tests/test_authority_change.py
 tests/test_baseline.py
 tests/test_baseline_compat.py
@@ -188,6 +198,7 @@ tests/test_env_dependency.py
 tests/test_escalation.py
 tests/test_escalation_remediation.py
 tests/test_evidence_contract_v7.py
+tests/test_evidence_metrics.py
 tests/test_evidence_type.py
 tests/test_exceptions.py
 tests/test_fingerprints.py
@@ -233,8 +244,10 @@ tests/test_scan_framework.py
 tests/test_scan_phase11.py
 tests/test_scorecard.py
 tests/test_scoring.py
+tests/test_security_brief.py
 tests/test_semantic.py
 tests/test_severity.py
+tests/test_subprocess_shell_detection.py
 tests/test_suppressions.py
 tests/test_target_taxonomy_engine.py
 tests/test_telemetry.py

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/SafeAI_Static_Analyzer.egg-info/requires.txt

@@ -3,7 +3,7 @@ PyYAML>=6.0.3
 [dev]
 pytest>=8.0
 pytest-cov>=5.0
-ruff<0.17,>=0.16.1
+ruff<0.18,>=0.16.1
 jinja2>=3.1.4
 jsonschema>=4.22.0
 typing-extensions>=4.4.0

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/pyproject.toml

@@ -34,7 +34,7 @@ dependencies = [
 dev = [
     "pytest>=8.0",
     "pytest-cov>=5.0",
-    "ruff>=0.16.1,<0.17",
+    "ruff>=0.16.1,<0.18",
     "jinja2>=3.1.4",
     "jsonschema>=4.22.0",
     "typing-extensions>=4.4.0",

safeai_static_analyzer-2.7.0/safeai/analysis/evidence_metrics.py

@@ -0,0 +1,188 @@
+"""Authority evidence metrics: how much of a scan's authority picture is
+evidence-backed, inferred, unattributed, or unknown.
+
+ROADMAP Layer 2 research metric and the P0 "unknown-rate surfacing" gap:
+the provenance fields SafeAI already emits are aggregated into a small,
+deterministic block attached to the JSON report and the manifest summary.
+
+Rules:
+  * Pure and deterministic — identical input yields identical output.
+  * Statement and finding metrics are computed from manifest-shaped data
+    only (``tool_surface`` entries and finding ``provenance_class``), so an
+    independent consumer can recompute them from a manifest without
+    SafeAI internals. Authority-dimension metrics need the baseline diff's
+    per-tool ``authority`` blocks (attributed tools plus the unattributed
+    entry) and are absent without a baseline.
+  * Rates follow the benchmark convention: an empty denominator is
+    ``insufficient_evidence`` with ``value: None`` — never 0, never 1.
+  * Unknown is counted, never reduced or reinterpreted. These metrics
+    describe evidence quality; no gate, policy, or score consumes them.
+"""
+
+from safeai.analysis.tool_identity import UNATTRIBUTED_KEY
+from safeai.kya.contract import PROVENANCE_CLASSES
+
+METRICS_VERSION = 1
+
+#: Per-tool authority dimensions carried by ``capability_diff`` tools
+#: (``authority_for`` in ``safeai/analysis/capability_diff.py``).
+AUTHORITY_DIMENSIONS = (
+    "destinations", "data_scope", "approval", "autonomy",
+    "delegation", "credential", "identity",
+)
+
+#: Measurements this block deliberately does not make, with the reason.
+NOT_MEASURED = (
+    {"metric": "false_attribution_rate",
+     "reason": "requires annotated ground truth; measured by `safeai benchmark`"},
+)
+
+
+def _ratio(numerator, denominator, reason):
+    if denominator <= 0:
+        return {"value": None, "status": "insufficient_evidence",
+                "numerator": 0, "denominator": 0, "reason": reason}
+    return {"value": round(numerator / denominator, 4), "status": "measured",
+            "numerator": numerator, "denominator": denominator}
+
+
+def _capability_statements(tool_surface):
+    """Yield ``(tool_key, capability)`` for every authority statement."""
+    for tool in tool_surface or []:
+        if not isinstance(tool, dict):
+            continue
+        key = str(tool.get("tool_key") or "")
+        for capability in tool.get("capabilities") or []:
+            if isinstance(capability, dict):
+                yield key, capability
+
+
+def _statement_counts(tool_surface):
+    counts = {"total": 0, "evidence_backed": 0, "inferred": 0,
+              "unattributed": 0, "with_evidence_refs": 0}
+    for key, capability in _capability_statements(tool_surface):
+        inferred = bool(capability.get("inferred"))
+        unattributed = key == UNATTRIBUTED_KEY
+        has_refs = bool(capability.get("evidence"))
+        counts["total"] += 1
+        counts["inferred"] += inferred
+        counts["unattributed"] += unattributed
+        counts["with_evidence_refs"] += has_refs
+        counts["evidence_backed"] += (not inferred and not unattributed and has_refs)
+    return counts
+
+
+def _finding_provenance(findings):
+    buckets = {cls: 0 for cls in PROVENANCE_CLASSES}
+    for finding in findings or []:
+        if not isinstance(finding, dict):
+            continue
+        cls = str(finding.get("provenance_class") or "unknown")
+        # An out-of-contract class is not evidence of anything: unknown.
+        buckets[cls if cls in buckets else "unknown"] += 1
+    return dict(sorted(buckets.items()))
+
+
+def _authority_dimensions(diff_tools):
+    """Unknown counts per authority dimension across diffed tools.
+
+    Returns ``None`` when no diffed tool carries an ``authority`` block
+    (no baseline comparison), so the metric is absent rather than empty.
+    """
+    by_dimension = {dim: {"total": 0, "unknown": 0} for dim in AUTHORITY_DIMENSIONS}
+    seen = False
+    for tool in diff_tools or []:
+        authority = tool.get("authority") if isinstance(tool, dict) else None
+        if not isinstance(authority, dict):
+            continue
+        seen = True
+        for dim in AUTHORITY_DIMENSIONS:
+            block = authority.get(dim)
+            provenance = block.get("provenance_class") if isinstance(block, dict) else None
+            by_dimension[dim]["total"] += 1
+            by_dimension[dim]["unknown"] += (provenance in (None, "unknown"))
+    return by_dimension if seen else None
+
+
+def _diffed_tools(capability_diff):
+    """Attributed tools plus the separately carried unattributed entry."""
+    if not isinstance(capability_diff, dict):
+        return []
+    tools = list(capability_diff.get("tools") or [])
+    if isinstance(capability_diff.get("unattributed"), dict):
+        tools.append(capability_diff["unattributed"])
+    return tools
+
+
+def compute_evidence_metrics(tool_surface, findings, capability_diff=None):
+    """Return the evidence-metrics block for one scan.
+
+    Parameters
+    ----------
+    tool_surface : list
+        Per-tool capability surface (report or manifest ``tool_surface``).
+    findings : list
+        Findings carrying ``provenance_class`` (report or manifest).
+    capability_diff : dict, optional
+        The report's ``capability_diff`` from a baseline comparison.
+    """
+    statements = _statement_counts(tool_surface)
+    provenance = _finding_provenance(findings)
+    dimensions = _authority_dimensions(_diffed_tools(capability_diff))
+    total = statements["total"]
+    no_statements = "no authority statements in this scan"
+
+    rates = {
+        "evidence_backed_authority_rate": _ratio(
+            statements["evidence_backed"], total, no_statements),
+        "inferred_authority_rate": _ratio(
+            statements["inferred"], total, no_statements),
+        "unattributed_authority_rate": _ratio(
+            statements["unattributed"], total, no_statements),
+        "evidence_completeness": _ratio(
+            statements["with_evidence_refs"], total, no_statements),
+        "unknown_finding_rate": _ratio(
+            provenance["unknown"], sum(provenance.values()),
+            "no findings in this scan"),
+    }
+    if dimensions is None:
+        rates["unknown_authority_dimension_rate"] = _ratio(
+            0, 0, "no baseline comparison in this scan")
+    else:
+        rates["unknown_authority_dimension_rate"] = _ratio(
+            sum(d["unknown"] for d in dimensions.values()),
+            sum(d["total"] for d in dimensions.values()),
+            "no diffed tools carry authority dimensions")
+
+    return {
+        "metrics_version": METRICS_VERSION,
+        "authority_statements": statements,
+        "finding_provenance": provenance,
+        "authority_dimensions": dimensions,
+        "rates": dict(sorted(rates.items())),
+        "not_measured": [dict(item) for item in NOT_MEASURED],
+    }
+
+
+def summary_lines(metrics):
+    """Short factual lines for human-facing reports (deterministic)."""
+    if not isinstance(metrics, dict):
+        return []
+    rates = metrics.get("rates") or {}
+
+    def fmt(name):
+        entry = rates.get(name) or {}
+        if entry.get("status") != "measured":
+            return "insufficient evidence"
+        return "{:.0%} ({}/{})".format(
+            entry["value"], entry["numerator"], entry["denominator"])
+
+    lines = [
+        "Evidence-backed authority: " + fmt("evidence_backed_authority_rate"),
+        "Inferred authority: " + fmt("inferred_authority_rate"),
+        "Unattributed authority: " + fmt("unattributed_authority_rate"),
+    ]
+    if (rates.get("unknown_authority_dimension_rate") or {}).get("status") == "measured":
+        lines.append("Unknown authority dimensions: "
+                     + fmt("unknown_authority_dimension_rate"))
+    return lines

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/safeai/analysis/import_graph.py

@@ -51,7 +51,7 @@ class ImportGraph:
     def resolve_module(self, module_name):
         return self.module_to_file.get(module_name)
 
-    def resolve_symbol(self, qualified_name):
+    def resolve_symbol(self, qualified_name, _visited=None):
         """Map a fully-qualified symbol to its defining module and file.
 
         Tries progressively shorter module prefixes (e.g. ``a.b.c.Var`` ->
@@ -59,6 +59,12 @@ class ImportGraph:
         chains when a symbol is re-exported through ``__init__.py``.
         Returns a dict with keys ``module``, ``file``, and ``symbol``, or ``None``.
         """
+        if _visited is None:
+            _visited = set()
+        if qualified_name in _visited:
+            return None
+        _visited.add(qualified_name)
+
         parts = qualified_name.split(".")
         for i in range(len(parts), 0, -1):
             module = ".".join(parts[:i])
@@ -70,7 +76,7 @@ class ImportGraph:
                 direct = symbol.split(".")[0]
                 target = self.symbol_index[path].get(direct)
                 if isinstance(target, dict) and target.get("kind") == "reexport":
-                    return self.resolve_symbol(target.get("target"))
+                    return self.resolve_symbol(target.get("target"), _visited)
             return {
                 "module": module,
                 "file": path,

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/safeai/analysis/semantic.py

@@ -78,14 +78,19 @@ def _literal_value(node):
     return None
 
 
-def _resolve_from_module(module, level, current_module):
+def _resolve_from_module(module, level, current_module, path=""):
     """Resolve a relative import to an absolute module name.
 
     ``level`` corresponds to the number of leading dots in a ``from ... import``.
+    When ``path`` ends with ``__init__.py`` the level is reduced by one because
+    a relative import from a package's ``__init__`` is resolved relative to the
+    package itself, not one level deeper.
     """
     if level == 0:
         return module or ""
     parts = current_module.split(".") if current_module else []
+    if path.endswith("__init__.py"):
+        level = max(0, level - 1)
     keep = max(0, len(parts) - level)
     prefix = ".".join(parts[:keep])
     if module:
@@ -112,7 +117,7 @@ def build_semantic_document(path, content, module_name=""):
                 local = alias.asname or alias.name.split(".")[0]
                 doc.imports[local] = alias.name
         elif isinstance(node, ast.ImportFrom):
-            base = _resolve_from_module(node.module, node.level, module_name)
+            base = _resolve_from_module(node.module, node.level, module_name, path)
             for alias in node.names:
                 local = alias.asname or alias.name
                 doc.from_imports[local] = f"{base}.{alias.name}" if base else alias.name

{safeai_static_analyzer-2.5.0 → safeai_static_analyzer-2.7.0}/safeai/analyzers/capability/analyzer.py

@@ -8,6 +8,7 @@ Operates in two passes:
      Also detects autonomous agent loop patterns.
 """
 
+import ast
 import re
 
 from safeai.analyzers import register_analyzer
@@ -23,6 +24,7 @@ CAP_PATTERNS = {
     "redis": re.compile(r"import redis|from redis|Redis\(|redis\.StrictRedis|redis\.Redis", re.IGNORECASE),
     "s3": re.compile(r"import boto3|from boto3|boto3\.client.*s3|boto3\.resource.*s3|S3Client|s3\.put|s3\.get", re.IGNORECASE),
     "slack": re.compile(r"import slack|from slack|SlackClient|slack_sdk|slack\.WebClient|slack\.SocketModeClient", re.IGNORECASE),
+    "ms_teams": re.compile(r"(?:\bimport\s+teams\b|\bfrom\s+teams\s+import\b|\bTeamsClient\b|\bteams_sdk\b|\bimport\s+pymsteams\b|\bfrom\s+pymsteams\b|\bpymsteams\.(?:connectorcard|cardsection)\b)", re.IGNORECASE),
     "jira": re.compile(r"import jira|from jira|JIRA\(|jira\.Client|jira\.JIRA", re.IGNORECASE),
     "browser_playwright": re.compile(r"import playwright|from playwright|Playwright", re.IGNORECASE),
     "browser_selenium": re.compile(r"import selenium|from selenium|webdriver|WebDriver", re.IGNORECASE),
@@ -32,6 +34,60 @@ CAP_PATTERNS = {
 
 # Higher-risk variants detected in addition to the base capability patterns.
 SUBPROCESS_SHELL_RE = re.compile(r"subprocess[\s\S]{0,200}?shell\s*=\s*True", re.IGNORECASE)
+
+# subprocess entry points that accept ``shell=``.
+SUBPROCESS_SHELL_FUNCS = {"run", "call", "check_call", "check_output", "Popen"}
+
+
+def _subprocess_shell_calls(content):
+    """Return ``(lineno, evidence)`` for subprocess calls passing ``shell=True``.
+
+    The per-line regex cannot see a call whose ``shell=True`` sits on a
+    later line than ``subprocess.run(`` — the usual black/ruff wrapping of
+    long calls. This AST pass resolves ``import subprocess [as x]`` and
+    ``from subprocess import run [as y]`` bindings and reports each call at
+    its first line. Only a literal ``True`` counts; ``**kwargs`` or
+    computed values stay unresolved. Unparseable sources yield nothing.
+    """
+    try:
+        tree = ast.parse(content)
+    except (SyntaxError, ValueError):
+        return []
+    modules, functions = set(), set()
+    for node in ast.walk(tree):
+        if isinstance(node, ast.Import):
+            for alias in node.names:
+                if alias.name == "subprocess":
+                    modules.add(alias.asname or "subprocess")
+        elif isinstance(node, ast.ImportFrom) and node.module == "subprocess":
+            for alias in node.names:
+                if alias.name in SUBPROCESS_SHELL_FUNCS:
+                    functions.add(alias.asname or alias.name)
+    if not modules and not functions:
+        return []
+
+    calls = []
+    for node in ast.walk(tree):
+        if not isinstance(node, ast.Call):
+            continue
+        func = node.func
+        is_subprocess = (
+            isinstance(func, ast.Attribute)
+            and func.attr in SUBPROCESS_SHELL_FUNCS
+            and isinstance(func.value, ast.Name)
+            and func.value.id in modules
+        ) or (isinstance(func, ast.Name) and func.id in functions)
+        if not is_subprocess:
+            continue
+        if any(
+            kw.arg == "shell"
+            and isinstance(kw.value, ast.Constant)
+            and kw.value.value is True
+            for kw in node.keywords
+        ):
+            segment = ast.get_source_segment(content, node) or ""
+            calls.append((node.lineno, " ".join(segment.split())[:200]))
+    return sorted(calls)
 FILE_WRITE_RE = re.compile(r"\bopen\([^)]*[\"'](?:w|a|x)[bt+]?[\"']", re.IGNORECASE)
 
 RULE_BY_CAP = {
@@ -45,6 +101,7 @@ RULE_BY_CAP = {
     "redis": "CAP_redis",
     "s3": "CAP_s3",
     "slack": "CAP_slack",
+    "ms_teams": "CAP_ms_teams",
     "jira": "CAP_jira",
     "browser_playwright": "CAP_browser_playwright",
     "browser_selenium": "CAP_browser_selenium",
@@ -63,6 +120,7 @@ CATEGORY_BY_CAP = {
     "redis": "Databases",
     "s3": "Cloud",
     "slack": "Collaboration",
+    "ms_teams": "Collaboration",
     "jira": "Collaboration",
     "browser_playwright": "Browser",
     "browser_selenium": "Browser",
@@ -268,6 +326,28 @@ class CapabilityAnalyzer:
                             "score_contribution": 12,
                             "remediation": "Introduce iteration limits and human approval checkpoints.",
                         })
+
+                # Multi-line subprocess calls the per-line regex missed.
+                # Single-line hits are already in ``seen`` (same key), so
+                # their line, evidence, and fingerprint are unchanged.
+                if path.endswith(".py"):
+                    for lineno, evidence in _subprocess_shell_calls(content):
+                        key = (path, "subprocess_shell", lineno)
+                        if key in seen:
+                            continue
+                        seen.add(key)
+                        rule = rule_map.get("CAP_subprocess_shell", {})
+                        findings.append(_finding(
+                            "CAP_subprocess_shell",
+                            rule,
+                            "subprocess invoked with shell=True",
+                            path,
+                            lineno,
+                            "Shell",
+                            evidence=evidence,
+                            confidence=0.7,
+                            score_contribution=15,
+                        ))
             except Exception:
                 continue
 

safeai_static_analyzer-2.7.0/safeai/benchmark/__init__.py

@@ -0,0 +1,21 @@
+"""Authority evidence validation benchmark (v2.6).
+
+Offline, deterministic corpus evaluation: runs the real ``safeai
+scan`` CLI path against ``tests/benchmarks/authority/`` and compares
+structurally with hand-verified expected truth. The benchmark
+measures; it never gates.
+"""
+
+from safeai.benchmark.runner import (
+    BenchmarkError,
+    discover_cases,
+    evaluate_case,
+    run_corpus,
+)
+
+__all__ = [
+    "BenchmarkError",
+    "discover_cases",
+    "evaluate_case",
+    "run_corpus",
+]