fix(agents): Impressum+Cookie delegieren MC-Laden ans Main Tool — Scope-Filter + Maßnahmen
CI / detect-changes (push) Successful in 8s
CI / branch-name (push) Has been skipped
CI / guardrail-integrity (push) Has been skipped
CI / secret-scan (push) Has been skipped
CI / sbom-scan (push) Has been skipped
CI / build-sha-integrity (push) Failing after 4s
CI / iace-gt-coverage (push) Has been skipped
CI / test-python-backend (push) Successful in 30s
CI / test-python-document-crawler (push) Has been skipped
CI / test-python-dsms-gateway (push) Has been skipped
CI / dep-audit (push) Has been skipped
CI / validate-canonical-controls (push) Successful in 11s
CI / loc-budget (push) Successful in 14s
CI / go-lint (push) Has been skipped
CI / python-lint (push) Has been skipped
CI / nodejs-lint (push) Has been skipped
CI / nodejs-build (push) Has been skipped
CI / test-go (push) Has been skipped

Regression: Der v3-Agent-Pfad baute eine parallele MC-Pipeline
(_load_impressum_mcs / _load_cookie_mcs, Roh-SELECT) und lief damit an
allen Schutzmechanismen der Engine vorbei → GOV/Branchen-MCs als HIGH bei
OEM/Zulieferer, fremde MCs (Bestellbestätigung), und action=check_question
(Fragen statt Maßnahmen im Frontend).

- Agent delegiert MC-Laden an rag_document_checker._load_controls
  (P72-Scope, check_type='text', fits_doc_type/scope_requires).
- Subtraktives Sektor-Gate (SECTOR_PREFIXES) + Themen-Gate am Agent-Rand.
- action = konkrete Maßnahme (Imperativ) statt check_question.
- rag_document_checker: from __future__ import annotations (3.9-Import).
- mcs: Name-Pattern erkennt "Aktiengesellschaft" (OEM-Impressums).
- Tote GT-/Semantic-/Routes-Tests wiederbelebt (v3-Mismatch +
  agent.cascade-Patch-Target). Alle 72 Specialist-Tests grün.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
2026-06-09 11:30:16 +02:00
parent bd4882e143
commit 389e6de0c7
13 changed files with 641 additions and 135 deletions
@@ -15,10 +15,17 @@ from compliance.services.specialist_agents import (
ImpressumAgent,
Severity,
)
from compliance.services.specialist_agents.impressum.agent import (
_build_measure,
)
from compliance.services.specialist_agents.impressum.regex_boost import (
BOOST_KEYWORDS,
boost_matches_db_mc,
compute_regex_boosts,
criteria_on_topic,
)
from compliance.services.specialist_agents.impressum.v3_engine import (
_filter_controls,
)
@@ -187,3 +194,152 @@ def test_short_text_skipped():
def test_agent_version_is_three():
agent = ImpressumAgent()
assert agent.agent_version == "3.0"
# ── Themen-Gate: criteria_on_topic ──────────────────────────────────
def test_criteria_on_topic_keeps_genuine_telefon():
assert criteria_on_topic([
"Telefonnummer angeben",
"Erreichbar per Telefon",
]) is True
def test_criteria_on_topic_keeps_genuine_address():
assert criteria_on_topic([
"Vollständige Postadresse (Straße, Hausnummer, PLZ, Ort)",
]) is True
def test_criteria_on_topic_drops_bestellbestaetigung():
# Fremd-MC: kein Impressum-Themenüberlapp → raus.
assert criteria_on_topic([
"Bestellbestätigung wird nach Vertragsschluss versendet",
"Bestelleingang wird dokumentiert",
]) is False
def test_criteria_on_topic_single_incidental_hit_dropped():
# 'E-Mail' allein (1 Treffer) reicht nicht — braucht >=2.
assert criteria_on_topic([
"Bestellbestätigung wird per E-Mail versendet",
]) is False
def test_criteria_on_topic_drops_behoerdliche_anzeige():
assert criteria_on_topic([
"Behördliche Anzeige der Tätigkeit erfolgt",
"Gewerbeanmeldung liegt vor",
]) is False
def test_criteria_on_topic_empty_kept():
# Keine Kriterien = kein Signal → konservativ behalten.
assert criteria_on_topic([]) is True
# ── Scope-Filter: _filter_controls ──────────────────────────────────
def _mc(control_id, pass_criteria):
return {"control_id": control_id, "pass_criteria": pass_criteria,
"fail_criteria": []}
def test_filter_controls_drops_gov_when_out_of_scope():
controls = [_mc("GOV-814-A03", ["Behörde meldet an Aufsichtsstelle"])]
kept, stats = _filter_controls(controls, business_scope=set())
assert kept == []
assert stats["sector_dropped"] == 1
def test_filter_controls_keeps_gov_when_in_scope():
controls = [_mc("GOV-814-A03",
["Aufsichtsbehörde und Behörde benannt"])]
kept, stats = _filter_controls(controls,
business_scope={"government"})
assert len(kept) == 1
assert stats["sector_dropped"] == 0
def test_filter_controls_keeps_genuine_impressum_mc():
controls = [_mc("AUTH-1954-A07",
["Vollständige Postadresse mit Straße und PLZ"])]
kept, stats = _filter_controls(controls, business_scope=set())
assert len(kept) == 1
assert stats["sector_dropped"] == 0
assert stats["offtopic_dropped"] == 0
def test_filter_controls_drops_offtopic_non_sector_mc():
controls = [_mc("ECOM-1-A1",
["Bestellbestätigung nach Vertragsschluss versenden"])]
kept, stats = _filter_controls(controls, business_scope=set())
assert kept == []
assert stats["offtopic_dropped"] == 1
# ── Maßnahme statt Frage: _build_measure ────────────────────────────
def test_build_measure_is_imperative_not_question():
m = _build_measure("USt-IdNr", "§ 5 Abs. 1 Nr. 6 TMG")
assert "?" not in m
assert "ergänzen" in m.lower()
assert "Rechtsgrundlage" in m
def test_build_measure_handles_empty_label():
m = _build_measure("", "")
assert "?" not in m
assert m.strip() != ""
# ── Delegation an Main-Tool-Engine + Filter (Integration) ───────────
def test_run_v3_pipeline_delegates_and_filters(monkeypatch):
"""run_v3_pipeline lädt über die Main-Tool-Engine (_load_controls
gemockt), normalisiert JSONB-Strings und das Sektor-/Themen-Gate
entfernt GOV (out-of-scope) + fremde MCs. Genuine MC bleibt."""
from compliance.services.specialist_agents.impressum import v3_engine
async def _fake_load(doc_type, db_url, limit, business_scope=None):
# pass_criteria absichtlich als JSON-STRING (wie asyncpg JSONB)
return [
{"control_id": "AUTH-1954-A07", "title": "USt-IdNr",
"regulation": "TMG", "article": "§ 5", "severity": "HIGH",
"check_question": "Ist die USt-IdNr angegeben?",
"pass_criteria": '["USt-IdNr"]',
"fail_criteria": "[]"},
{"control_id": "GOV-814-A03", "title": "Behördliche Anzeige",
"regulation": "X", "article": "", "severity": "HIGH",
"check_question": "Behörde informiert?",
"pass_criteria": '["Aufsichtsbehörde und Behörde benannt"]',
"fail_criteria": "[]"},
{"control_id": "ECOM-1-A1", "title": "Bestellbestätigung",
"regulation": "X", "article": "", "severity": "HIGH",
"check_question": "Bestellbestätigung versandt?",
"pass_criteria":
'["Bestellbestätigung nach Vertragsschluss versenden"]',
"fail_criteria": "[]"},
]
monkeypatch.setattr(
"compliance.services.rag_document_checker._load_controls",
_fake_load,
)
# AUTH-MC matched per Keyword → kein Layer-2-Embedding nötig; kein
# mc_embedding_matcher-Mock erforderlich.
text = ("Beispiel GmbH\nMusterstr. 1\n12345 Berlin\n"
"USt-IdNr: DE123456789\n") * 5 # >100 Zeichen
results, telem = _run(
v3_engine.run_v3_pipeline(text, business_scope=set()),
)
cids = {r["control_id"] for r in results}
assert "GOV-814-A03" not in cids # Sektor out-of-scope
assert "ECOM-1-A1" not in cids # themenfremd
assert "AUTH-1954-A07" in cids # genuine MC bleibt
assert telem["sector_dropped"] == 1
assert telem["offtopic_dropped"] == 1