@cohortapp/agent-sdk 2.3.0

package/scripts/test-information-barriers.py

@@ -0,0 +1,484 @@
+#!/usr/bin/env python3
+"""Test suite for the Information Barrier System — Layers 1 and 2.
+
+Verifies:
+  1. Module imports
+  2. Disclosure boundary generation (profiles, conflicts, excluded domains)
+  3. Disclosure assessment — basic cases
+  4. Disclosure assessment — with provenance
+  5. Severity routing constants
+  6. Boundary generation with provenance
+  7. Audit logging
+
+Usage:
+  python3 scripts/test-information-barriers.py
+
+Created: 2026-04-05 — Task 10, Information Barrier System
+"""
+
+import json
+import os
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+
+SCRIPT_DIR = Path(__file__).resolve().parent
+REPO_ROOT = SCRIPT_DIR.parent
+PROVENANCE_INDEX = REPO_ROOT / "memory" / "provenance" / "index.jsonl"
+AUDIT_LOG_DIR = REPO_ROOT / "logs" / "audit"
+
+# Ensure scripts dir is on path
+sys.path.insert(0, str(SCRIPT_DIR))
+
+passed = 0
+failed = 0
+total = 0
+
+
+def test(name, condition, detail=""):
+    global passed, failed, total
+    total += 1
+    if condition:
+        passed += 1
+        print(f"  PASS: {name}")
+    else:
+        failed += 1
+        print(f"  FAIL: {name}" + (f" — {detail}" if detail else ""))
+
+
+def section(title):
+    print(f"\n{'='*60}")
+    print(f"  {title}")
+    print(f"{'='*60}")
+
+
+# ─────────────────────────────────────────────────────
+# Section 1: Module Imports
+# ─────────────────────────────────────────────────────
+section("1. Module Imports")
+
+try:
+    from disclosure_boundaries import generate_disclosure_boundaries
+    test("disclosure_boundaries imports successfully", True)
+except ImportError as e:
+    test("disclosure_boundaries imports successfully", False, str(e))
+    print("\nFATAL: Cannot continue without disclosure_boundaries. Exiting.")
+    sys.exit(1)
+
+try:
+    from disclosure_assessment import assess_disclosure
+    test("disclosure_assessment imports successfully", True)
+except ImportError as e:
+    test("disclosure_assessment imports successfully", False, str(e))
+    print("\nFATAL: Cannot continue without disclosure_assessment. Exiting.")
+    sys.exit(1)
+
+
+# ─────────────────────────────────────────────────────
+# Section 2: Disclosure Boundary Generation
+# ─────────────────────────────────────────────────────
+section("2. Disclosure Boundary Generation")
+
+# --- Casey ---
+shayan = generate_disclosure_boundaries("casey-morgan")
+test("Casey: boundaries not None", shayan is not None)
+
+if shayan is not None:
+    active_conflicts = shayan.get("active_conflicts") or []
+    excluded_domains = shayan.get("excluded_domains_active") or []
+    summary = shayan.get("recipient_clearance_summary") or ""
+
+    test(
+        "Casey: has 2 active conflicts (JVA + Seed SAFE)",
+        len(active_conflicts) == 2,
+        f"found {len(active_conflicts)} conflicts",
+    )
+
+    test(
+        "Casey: has 7+ excluded domains",
+        len(excluded_domains) >= 7,
+        f"found {len(excluded_domains)} excluded domains",
+    )
+
+    summary_lower = summary.lower()
+    test(
+        "Casey: summary mentions partner/JV/PartnerCo",
+        any(word in summary_lower for word in ("partner", "jv", "partnerco")),
+        f"summary: {summary[:120]}",
+    )
+
+    conflict_matters = [c.get("matter", "") for c in active_conflicts if isinstance(c, dict)]
+    test(
+        "Casey: conflict matter jva-review-northwind-group-investments present",
+        "jva-review-northwind-group-investments" in conflict_matters,
+        f"found matters: {conflict_matters}",
+    )
+    test(
+        "Casey: conflict matter seed-safe-review present",
+        "seed-safe-review" in conflict_matters,
+        f"found matters: {conflict_matters}",
+    )
+
+    excluded_set = {str(d).lower() for d in excluded_domains}
+    for domain in ("internal-legal", "financial", "regulatory-strategy"):
+        test(
+            f"Casey: excluded domain '{domain}' present",
+            domain in excluded_set,
+            f"excluded_domains_active: {excluded_domains}",
+        )
+else:
+    # Skip sub-tests
+    for _ in range(8):
+        test("(skipped — Casey profile not loaded)", False, "profile returned None")
+
+# --- Alex ---
+alex = generate_disclosure_boundaries("alex-chen")
+test("Alex: boundaries not None", alex is not None)
+
+if alex is not None:
+    alex_summary = (alex.get("recipient_clearance_summary") or "").lower()
+    test(
+        "Alex: unrestricted (summary contains 'unrestricted')",
+        "unrestricted" in alex_summary,
+        f"summary: {alex_summary[:120]}",
+    )
+    test(
+        "Alex: 0 conflicts",
+        len(alex.get("active_conflicts") or []) == 0,
+        f"conflicts: {alex.get('active_conflicts')}",
+    )
+    test(
+        "Alex: 0 excluded domains",
+        len(alex.get("excluded_domains_active") or []) == 0,
+        f"excluded: {alex.get('excluded_domains_active')}",
+    )
+else:
+    for _ in range(3):
+        test("(skipped — Alex profile not loaded)", False, "profile returned None")
+
+# --- Jordan ---
+nima = generate_disclosure_boundaries("jordan-blake")
+test("Jordan: boundaries not None", nima is not None)
+
+if nima is not None:
+    test(
+        "Jordan: 0 conflicts",
+        len(nima.get("active_conflicts") or []) == 0,
+        f"conflicts: {nima.get('active_conflicts')}",
+    )
+    nima_excluded = nima.get("excluded_domains_active") or []
+    nima_excluded_lower = [str(d).lower() for d in nima_excluded]
+    test(
+        "Jordan: has excluded domains including engineering and product",
+        "engineering" in nima_excluded_lower or "product" in nima_excluded_lower,
+        f"excluded_domains_active: {nima_excluded}",
+    )
+else:
+    for _ in range(2):
+        test("(skipped — Jordan profile not loaded)", False, "profile returned None")
+
+# --- Unknown person ---
+unknown = generate_disclosure_boundaries("totally-unknown-person-xyz")
+test("Unknown person: returns None", unknown is None, f"got: {unknown}")
+
+
+# ─────────────────────────────────────────────────────
+# Section 3: Disclosure Assessment — Basic Cases
+# ─────────────────────────────────────────────────────
+section("3. Disclosure Assessment — Basic Cases")
+
+# Clean message to Casey (no provenance)
+result_clean = assess_disclosure(
+    draft="Hi Casey, hope you're well. Any update on the PartnerCo ops calendar?",
+    recipient_slug="casey-morgan",
+)
+test(
+    "Clean message to Casey (no provenance): outcome is 'passed'",
+    result_clean.get("outcome") == "passed",
+    f"outcome: {result_clean.get('outcome')}, severity: {result_clean.get('severity')}",
+)
+
+# Empty draft to Casey
+result_empty = assess_disclosure(
+    draft="",
+    recipient_slug="casey-morgan",
+)
+test(
+    "Empty draft to Casey: outcome is 'passed'",
+    result_empty.get("outcome") == "passed",
+    f"outcome: {result_empty.get('outcome')}",
+)
+
+# Message to Alex (unrestricted)
+result_alex = assess_disclosure(
+    draft="The JVA review session with Jordan covered sensitive governance terms. SAFE at $50M.",
+    recipient_slug="alex-chen",
+)
+test(
+    "Message to Alex (unrestricted): always 'passed'",
+    result_alex.get("outcome") == "passed",
+    f"outcome: {result_alex.get('outcome')}, severity: {result_alex.get('severity')}",
+)
+
+
+# ─────────────────────────────────────────────────────
+# Section 4: Disclosure Assessment — With Provenance
+# ─────────────────────────────────────────────────────
+section("4. Disclosure Assessment — With Provenance")
+
+test_provenance = [
+    {
+        "source_id": "meeting-2026-04-04-jva-review",
+        "source_type": "meeting",
+        "participants": ["alex-chen", "jordan-blake"],
+        "originator": "jordan-blake",
+        "context_domain": "internal-legal",
+        "matter": "jva-review-northwind-group-investments",
+        "sensitivity_signals": ["legal_privilege", "negotiation_position"],
+        "timestamp": datetime.now(timezone.utc).isoformat(),
+        "classification_confidence": "high",
+    },
+    {
+        "source_id": "meeting-2026-04-04-partnerco-ops",
+        "source_type": "meeting",
+        "participants": ["alex-chen", "casey-morgan"],
+        "originator": "casey-morgan",
+        "context_domain": "partner-operations",
+        "matter": "partner-operations",
+        "sensitivity_signals": [],
+        "timestamp": datetime.now(timezone.utc).isoformat(),
+        "classification_confidence": "high",
+    },
+    {
+        "source_id": "email-2026-04-04-seed-safe",
+        "source_type": "email",
+        "participants": ["alex-chen", "jordan-blake"],
+        "originator": "jordan-blake",
+        "context_domain": "financial",
+        "matter": "seed-safe-review",
+        "sensitivity_signals": ["financial_terms", "negotiation_position"],
+        "timestamp": datetime.now(timezone.utc).isoformat(),
+        "classification_confidence": "high",
+    },
+]
+
+# Save original provenance content
+original_provenance = None
+if PROVENANCE_INDEX.exists():
+    original_provenance = PROVENANCE_INDEX.read_bytes()
+
+try:
+    # Write test provenance records
+    PROVENANCE_INDEX.parent.mkdir(parents=True, exist_ok=True)
+    with PROVENANCE_INDEX.open("w", encoding="utf-8") as fh:
+        for rec in test_provenance:
+            fh.write(json.dumps(rec) + "\n")
+
+    # Scenario 1: Casey receives message referencing JVA review
+    result_shayan_jva = assess_disclosure(
+        draft="Following the JVA review session, we're looking at adjusting the governance terms",
+        recipient_slug="casey-morgan",
+    )
+    test(
+        "Casey JVA incident: draft is blocked (high or critical severity)",
+        result_shayan_jva.get("outcome") == "blocked"
+        and result_shayan_jva.get("severity") in ("critical", "high"),
+        f"outcome: {result_shayan_jva.get('outcome')}, severity: {result_shayan_jva.get('severity')}",
+    )
+
+    # Scenario 2: Casey receives message referencing Seed SAFE
+    result_shayan_safe = assess_disclosure(
+        draft="The seed safe review is progressing, we need to adjust the valuation terms",
+        recipient_slug="casey-morgan",
+    )
+    test(
+        "Casey Seed SAFE: draft is blocked",
+        result_shayan_safe.get("outcome") == "blocked",
+        f"outcome: {result_shayan_safe.get('outcome')}, severity: {result_shayan_safe.get('severity')}",
+    )
+
+    # Scenario 3: Casey receives message about PartnerCo ops (he was a participant)
+    result_shayan_partnerco = assess_disclosure(
+        draft="Hi Casey, the PartnerCo ops meeting went well",
+        recipient_slug="casey-morgan",
+    )
+    test(
+        "Casey PartnerCo ops: passed or low severity (he was participant)",
+        result_shayan_partnerco.get("outcome") == "passed"
+        or result_shayan_partnerco.get("severity") in (None, "low"),
+        f"outcome: {result_shayan_partnerco.get('outcome')}, severity: {result_shayan_partnerco.get('severity')}",
+    )
+
+    # Scenario 4: Jordan receives message referencing JVA review.
+    # Although Jordan was a participant, the provenance record carries legal_privilege
+    # and negotiation_position signals — these trigger the hard privilege gate
+    # regardless of participant status. The engine correctly blocks to prevent
+    # inadvertent disclosure via drafts that happen to reference the record.
+    result_nima_jva = assess_disclosure(
+        draft="Following the JVA review session, let's finalize the governance terms",
+        recipient_slug="jordan-blake",
+    )
+    test(
+        "Jordan JVA content: privilege gate fires (blocked) — participant status does not bypass privilege signals",
+        result_nima_jva.get("outcome") == "blocked"
+        and result_nima_jva.get("severity") in ("critical", "high"),
+        f"outcome: {result_nima_jva.get('outcome')}, severity: {result_nima_jva.get('severity')}",
+    )
+
+    # Scenario 5: Alex receives very sensitive content — always passes
+    result_alex_sensitive = assess_disclosure(
+        draft="JVA review: Jordan recommends better terms. Seed SAFE at $50M.",
+        recipient_slug="alex-chen",
+    )
+    test(
+        "Alex sensitive content: always passed",
+        result_alex_sensitive.get("outcome") == "passed",
+        f"outcome: {result_alex_sensitive.get('outcome')}, severity: {result_alex_sensitive.get('severity')}",
+    )
+
+    # ─────────────────────────────────────────────────────
+    # Section 6: Boundary Generation with Provenance
+    # ─────────────────────────────────────────────────────
+    section("6. Boundary Generation with Provenance")
+
+    shayan_with_prov = generate_disclosure_boundaries("casey-morgan")
+    test(
+        "Casey boundaries not None with provenance loaded",
+        shayan_with_prov is not None,
+    )
+
+    if shayan_with_prov is not None:
+        recent_ctx = shayan_with_prov.get("recent_sensitive_context") or []
+        test(
+            "Casey: recent_sensitive_context has >= 1 entry",
+            len(recent_ctx) >= 1,
+            f"recent_sensitive_context count: {len(recent_ctx)}",
+        )
+
+        # Check that JVA meeting is flagged somewhere in the context
+        all_sources_str = json.dumps(recent_ctx).lower()
+        test(
+            "Casey: JVA meeting flagged in recent_sensitive_context",
+            "jva" in all_sources_str,
+            f"sources: {all_sources_str[:200]}",
+        )
+    else:
+        for _ in range(2):
+            test("(skipped — Casey boundaries returned None)", False, "profile returned None")
+
+finally:
+    # Always restore the original provenance file
+    if original_provenance is not None:
+        PROVENANCE_INDEX.write_bytes(original_provenance)
+    else:
+        # There was no original — truncate back to empty
+        if PROVENANCE_INDEX.exists():
+            PROVENANCE_INDEX.write_text("", encoding="utf-8")
+
+
+# ─────────────────────────────────────────────────────
+# Section 5: Severity Routing
+# ─────────────────────────────────────────────────────
+section("5. Severity Routing")
+
+from disclosure_assessment import SEVERITY_ACTIONS, SEVERITY_ORDER
+
+test(
+    "SEVERITY_ACTIONS: critical → block_and_escalate",
+    SEVERITY_ACTIONS.get("critical") == "block_and_escalate",
+    f"got: {SEVERITY_ACTIONS.get('critical')}",
+)
+test(
+    "SEVERITY_ACTIONS: high → block_and_flag",
+    SEVERITY_ACTIONS.get("high") == "block_and_flag",
+    f"got: {SEVERITY_ACTIONS.get('high')}",
+)
+test(
+    "SEVERITY_ACTIONS: medium → strip_and_send",
+    SEVERITY_ACTIONS.get("medium") == "strip_and_send",
+    f"got: {SEVERITY_ACTIONS.get('medium')}",
+)
+test(
+    "SEVERITY_ACTIONS: low → warn_and_send",
+    SEVERITY_ACTIONS.get("low") == "warn_and_send",
+    f"got: {SEVERITY_ACTIONS.get('low')}",
+)
+
+test(
+    "SEVERITY_ORDER: critical(3) > high(2) > medium(1) > low(0)",
+    SEVERITY_ORDER.get("critical") == 3
+    and SEVERITY_ORDER.get("high") == 2
+    and SEVERITY_ORDER.get("medium") == 1
+    and SEVERITY_ORDER.get("low") == 0,
+    f"got: {SEVERITY_ORDER}",
+)
+
+
+# ─────────────────────────────────────────────────────
+# Section 7: Audit Logging
+# ─────────────────────────────────────────────────────
+section("7. Audit Logging")
+
+# Run an assessment to trigger a log write (provenance is now restored/empty so
+# any keyword hits are against the real index, which may be empty — that's fine;
+# we just need assess_disclosure to run and log).
+assess_disclosure(
+    draft="Audit log test — checking that disclosure assessment writes an entry.",
+    recipient_slug="casey-morgan",
+)
+
+date_str = datetime.now(timezone.utc).strftime("%Y-%m-%d")
+log_path = AUDIT_LOG_DIR / f"{date_str}-disclosure-assessments.jsonl"
+
+test(
+    f"Audit log file exists at logs/audit/{date_str}-disclosure-assessments.jsonl",
+    log_path.exists(),
+    f"path checked: {log_path}",
+)
+
+if log_path.exists():
+    lines = [l for l in log_path.read_text(encoding="utf-8").strip().split("\n") if l.strip()]
+    test("Audit log has at least one entry", len(lines) >= 1, f"lines: {len(lines)}")
+
+    if lines:
+        try:
+            last_entry = json.loads(lines[-1])
+        except json.JSONDecodeError as e:
+            last_entry = {}
+            test("Last audit entry is valid JSON", False, str(e))
+        else:
+            test("Last audit entry is valid JSON", True)
+
+        test(
+            "Last audit entry has 'timestamp'",
+            "timestamp" in last_entry,
+            f"keys: {list(last_entry.keys())}",
+        )
+        test(
+            "Last audit entry has 'outcome'",
+            "outcome" in last_entry,
+            f"keys: {list(last_entry.keys())}",
+        )
+        test(
+            "Last audit entry has 'recipients'",
+            "recipients" in last_entry,
+            f"keys: {list(last_entry.keys())}",
+        )
+        test(
+            "Last audit entry has 'draft_hash'",
+            "draft_hash" in last_entry,
+            f"keys: {list(last_entry.keys())}",
+        )
+else:
+    for _ in range(5):
+        test("(skipped — audit log file not found)", False, "file absent")
+
+
+# ─────────────────────────────────────────────────────
+# Summary
+# ─────────────────────────────────────────────────────
+print(f"\n{'='*60}")
+print(f"  RESULTS: {passed}/{total} passed, {failed} failed")
+print(f"{'='*60}\n")
+
+sys.exit(0 if failed == 0 else 1)