feat(audit-pipeline): P72-v2 Heuristik nachgeschaerft + P80 Mini-Replay-Endpoint
CI / detect-changes (push) Successful in 9s
CI / branch-name (push) Has been skipped
CI / guardrail-integrity (push) Has been skipped
CI / nodejs-lint (push) Has been skipped
CI / secret-scan (push) Has been skipped
CI / dep-audit (push) Has been skipped
CI / sbom-scan (push) Has been skipped
CI / validate-canonical-controls (push) Successful in 13s
CI / loc-budget (push) Failing after 14s
CI / go-lint (push) Has been skipped
CI / python-lint (push) Has been skipped
CI / test-go (push) Has been skipped
CI / iace-gt-coverage (push) Has been skipped
CI / test-python-backend (push) Successful in 36s
CI / test-python-document-crawler (push) Has been skipped
CI / test-python-dsms-gateway (push) Has been skipped
CI / nodejs-build (push) Has been skipped

P72-v2  MC-Scope-Classifier Heuristik v2 — v1 hatte 79% 'other'-Bucket
        (Patterns zu strict). v2 deckt deutlich breiter ab:
          - DSE: Art. 13/14 + Betroffenenrechte (Art. 15-22) + DSB +
            Aufsichtsbehoerde + Speicherdauer + besondere Kategorien
          - TOM: Art. 32 + Verschluesselung/Backup/Pseudonymisierung +
            Zugriffskontrolle + ISO 27001 + BSI-Grundschutz + Audit-Log
          - cookie_richtlinie: Tracking-Pixel + Webstorage + GA/Matomo/
            Hotjar/Pixel/GTM
          - process: VVT (Art. 30) + DSFA (Art. 35) + Datenpannen
            (Art. 33/34) + HinSchG + Schulungen + Loeschkonzept
        Script `backfill_mc_scope_v2.py` re-classifiziert NUR den
        'other'-Bucket (spezifische v1-Buckets bleiben unangetastet).

P80    Mini-Replay-Endpoint (v1):
          POST /compliance-check/snapshots/{id}/replay
          ?recipient=foo@bar.com & dry_run=false
        Laedt Snapshot, rendert Mail mit AKTUELLEM Render-Code (P63-P67,
        P59b/P61/P62). Sendet [REPLAY]-prefixed Mail oder gibt nur
        HTML-Stats zurueck (dry_run).
        Effekt: 7min Re-Scan -> 2-5sec fuer Mail-Layout-Iterationen.
        v2 (spaeter): MC-Scorecard mit aktuellem scope_doc_type-Filter
        ueber Snapshot — erfordert _run_compliance_check Refactoring.

Plus Bugfix: GET /snapshots/{id} raised jetzt HTTPException statt
Tuple-Return (FastAPI hat Tuple als JSON-Array zurueckgegeben).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
2026-05-21 10:21:56 +02:00
parent cde670617e
commit 4946571863
3 changed files with 431 additions and 1 deletions
@@ -190,18 +190,44 @@ async def list_snapshots(domain: str = "", limit: int = 20):
@router.get("/snapshots/{snapshot_id}")
async def get_snapshot(snapshot_id: str):
"""P80: load full snapshot raw data."""
from fastapi import HTTPException
from database import SessionLocal
from compliance.services.check_snapshot import load_snapshot
db = SessionLocal()
try:
snap = load_snapshot(db, snapshot_id)
if not snap:
return {"error": "snapshot not found"}, 404
raise HTTPException(status_code=404, detail="snapshot not found")
return snap
finally:
db.close()
@router.post("/snapshots/{snapshot_id}/replay")
async def replay_snapshot(
snapshot_id: str,
recipient: str = "",
dry_run: bool = True,
):
"""P80: replay audit mail render from snapshot. 7min->2sec test cycle.
Default dry_run=true just returns rendered HTML size + section breakdown.
Pass recipient + dry_run=false to actually send a [REPLAY] mail.
"""
from database import SessionLocal
from compliance.services.check_replay import replay_from_snapshot
db = SessionLocal()
try:
return replay_from_snapshot(
db,
snapshot_id=snapshot_id,
recipient=(recipient if recipient else None),
dry_run=dry_run,
)
finally:
db.close()
async def _run_compliance_check(check_id: str, req: ComplianceCheckRequest):
"""Background task: check all documents with business-profile context."""
try:
@@ -0,0 +1,147 @@
"""
P80 — Replay-Pipeline (Mini-Version v1).
Lädt einen persistierten Snapshot und rendert die Audit-Mail mit dem
AKTUELLEN Mail-Render-Code neu. Nutzbar fuer:
* Mail-Layout-Aenderungen (P63-P67, P82 1-Pager, P84 Diff-Mode) testen
* Action-Recipes anpassen
* Disclaimer-Text iterieren
* Pattern-Notice-Logik tunen
NICHT enthalten (kommt in v2):
* MC-Scorecard re-run mit aktuellem scope_doc_type-Filter (P72) —
erfordert MC-Pipeline-Refactoring aus _run_compliance_check
* Vendor-Redundancy-Analyse re-run
Effekt v1: 7min Re-Scan -> 2-5 Sek fuer Mail-Layout-Iterationen.
Effekt v2 (spaeter): auch fuer MC-Filter-Tests.
"""
from __future__ import annotations
import logging
from typing import Any
from sqlalchemy.orm import Session
from compliance.services.check_snapshot import load_snapshot
logger = logging.getLogger(__name__)
def replay_from_snapshot(
db: Session,
snapshot_id: str,
recipient: str | None = None,
dry_run: bool = False,
) -> dict:
"""Replay audit mail render from snapshot.
Args:
db: SQLAlchemy session
snapshot_id: UUID of snapshot to replay
recipient: Override email recipient. None = skip send.
dry_run: If True, render HTML but do not send mail.
Returns:
{"snapshot_id", "html_size", "sections", "mail_sent", "preview"}
"""
snap = load_snapshot(db, snapshot_id)
if not snap:
return {"error": "snapshot not found", "snapshot_id": snapshot_id}
doc_entries = snap.get("doc_entries") or []
banner_result = snap.get("banner_result") or {}
profile_dict = snap.get("profile") or {}
cmp_vendors = snap.get("cmp_vendors") or []
site_label = snap.get("site_label") or snap.get("site_domain")
# Reconstruct doc_texts mapping (was the input to mail-render)
doc_texts: dict[str, str] = {}
for e in doc_entries:
dt = e.get("doc_type", "")
txt = (e.get("full_text") or e.get("text_preview") or "").strip()
if dt and txt:
doc_texts[dt] = txt
# Build results list mock (just enough for mail-render)
from compliance.services.doc_checks.runner import DocCheckResult
def _dict_to_result(d: dict) -> Any:
"""Best-effort reconstruction. Snapshot didn't persist DocCheckResult
so we fake minimal fields. For real MC-replay (v2) we'd re-run the
check_document_completeness function against the snapshot text."""
return type("R", (), {
"doc_type": d.get("doc_type", "other"),
"label": d.get("doc_type", "Dokument"),
"completeness_pct": d.get("completeness_pct", 0),
"correctness_pct": d.get("correctness_pct"),
"checks": [],
"error": d.get("error", ""),
})()
results = [_dict_to_result(e) for e in doc_entries]
# Render mail sections
section_sizes: dict[str, int] = {}
parts: list[str] = []
try:
from compliance.api.agent_doc_check_critical import build_critical_findings_html
critical_html = build_critical_findings_html(banner_result, None, results) or ""
parts.append(critical_html)
section_sizes["critical"] = len(critical_html)
except Exception as e:
logger.warning("Replay: critical-block failed: %s", e)
try:
from compliance.api.scope_disclaimer import build_scope_disclaimer_html
disclaimer = build_scope_disclaimer_html()
parts.append(disclaimer)
section_sizes["disclaimer"] = len(disclaimer)
except Exception as e:
logger.warning("Replay: disclaimer failed: %s", e)
try:
from compliance.api.agent_doc_check_banner import build_banner_deep_html
banner_html = build_banner_deep_html(banner_result) or ""
parts.append(banner_html)
section_sizes["banner"] = len(banner_html)
except Exception as e:
logger.warning("Replay: banner-block failed: %s", e)
try:
from compliance.api.vvt_table_renderer import build_vvt_table_html
vvt_html = build_vvt_table_html(cmp_vendors) or ""
parts.append(vvt_html)
section_sizes["vvt"] = len(vvt_html)
except Exception as e:
logger.warning("Replay: vvt failed: %s", e)
full_html = "".join(parts)
result = {
"snapshot_id": snapshot_id,
"check_id": snap.get("check_id"),
"site_domain": snap.get("site_domain"),
"html_size": len(full_html),
"sections": section_sizes,
"mail_sent": False,
"preview": full_html[:500] + "..." if len(full_html) > 500 else full_html,
}
if recipient and not dry_run:
try:
from compliance.services.email_sender import send_email
email_res = send_email(
recipient=recipient,
subject=f"[REPLAY] {site_label} (Snapshot {snapshot_id[:8]})",
body_html=full_html,
)
result["mail_sent"] = (email_res.get("status") == "sent")
result["mail_status"] = email_res.get("status")
except Exception as e:
logger.warning("Replay: mail send failed: %s", e)
result["mail_send_error"] = str(e)[:200]
return result
@@ -0,0 +1,257 @@
#!/usr/bin/env python3
"""P72 v2 — Heuristik nachschaerfen.
v1 hatte 79% 'other'-Bucket. v2 erweitert Patterns deutlich:
* DSE: deckt allgemeine Art. 13/14 + Betroffenenrechte ab
* TOM: deckt Art. 32 + technische Sicherheit breit ab
* cookie_richtlinie: alle Tracking/Analyse-Patterns
* process: alle Schulung/Verfahren/Meldepflicht-Patterns
Re-classify NUR den 'other'-Bucket — die anderen Buckets aus v1 sind bereits
spezifisch genug, die wollen wir nicht nochmal anfassen.
"""
from __future__ import annotations
import os
import re
import sys
from typing import Pattern
import psycopg2
# Patterns v2: deutlich lockerer (vorher zu strict, 79% landeten in 'other')
RULES_V2: list[tuple[str, Pattern]] = [
# Banner-UI (sehr eng — bleibt v1)
(
"banner_implementation",
re.compile(
r"\b(banner|cookie[-\s]?wall|pre[-\s]?ticked|"
r"vorausgewaehlt(e|en)?\s*checkbox|"
r"browser[-\s]?(default|standard|einstellung).{0,30}einwilligung|"
r"dark[-\s]?pattern|reject.{0,20}button|ablehn.{0,20}button|"
r"floating.{0,20}icon|cookie[-\s]?einstellungen)"
),
),
# CMP-Audit
(
"cmp_audit",
re.compile(
r"\b(consent[-\s]?(log|trail|audit)|"
r"einwilligungs(nachweis|log|trail|protokoll)|"
r"datensaetze?.{0,30}einwilligung|"
r"zeitstempel.{0,30}einwilligung)"
),
),
# AVV (Art. 28)
(
"avv",
re.compile(
r"\b(art\.?\s*28|auftragsverarbeit|adv|avv|"
r"data[-\s]?processing[-\s]?agreement|dpa|"
r"weisungsgebunden(er)?\s*auftragsverarbeit)"
),
),
# JC (Art. 26)
(
"jc",
re.compile(
r"\b(art\.?\s*26|joint[-\s]?controller|"
r"gemeinsam(e|er)\s*verantwortlich|"
r"konzern.{0,40}(verantwortlich|verarbeit)|"
r"gemeinsame.{0,20}verarbeitung)"
),
),
# Impressum
(
"impressum",
re.compile(
r"\b((paragraph|§)\s*5\s*(tmg|ddg)|"
r"§\s*18\s*mstv|"
r"impressum|anbieterkennzeichnung|"
r"geschaeftsbrief|firma.{0,20}kaufmann|"
r"vollstaendige.{0,20}geschaeftsadresse|"
r"postalische?.{0,30}adresse|"
r"handelsregister.{0,30}eintrag|"
r"§\s*55\s*rstv)"
),
),
# AGB
(
"agb",
re.compile(
r"\b(agb|allgemeine\s*geschaeftsbedingungen|"
r"vertragsbedingungen|"
r"§\s*305.{0,5}(bgb)?|"
r"klausel.{0,30}wirksam|"
r"vertragsabschluss.{0,40}online)"
),
),
# Widerruf
(
"widerruf",
re.compile(
r"\b(widerrufsbelehrung|widerrufsrecht|"
r"14.{0,10}tage.{0,10}frist|"
r"musterwiderruf|"
r"§\s*355\s*bgb)"
),
),
# Accounting
(
"accounting",
re.compile(
r"\b((rechnung|invoice).{0,30}(angeben|enthalten|fuehren)|"
r"§\s*14\s*ustg|umsatzsteueridentifikation\s+nummer.{0,30}rechnung|"
r"buchhaltung|"
r"steuernummer.{0,30}rechnung)"
),
),
# Cookie-Richtlinie (erweitert)
(
"cookie_richtlinie",
re.compile(
r"\b(cookie[-\s]?(richtlinie|policy|liste|tabelle|verzeichnis)|"
r"§\s*25\s*(tddg|tdddg|ttdsg)|"
r"§\s*165.{0,5}tkg|"
r"tracking[-\s]?technologi|"
r"tracking[-\s]?(pixel|skript)|"
r"webstorage|local[-\s]?storage.{0,30}einwilligung|"
r"cookie.{0,30}(zweck|speicherdauer|drittland|anbieter)|"
r"google\s+analytics|matomo|piwik|hotjar|"
r"facebook\s+pixel|meta\s+pixel|google\s+tag\s+manager)"
),
),
# TOM (deutlich erweitert)
(
"tom",
re.compile(
r"\b(art\.?\s*32|"
r"verschluesselung|encryption|kryptograph|"
r"backup|wiederherstell|recovery|"
r"pseudonymisier|anonymisier|"
r"zugriffskontrolle|berechtigungskonzept|"
r"benutzerverwaltung|identity[-\s]?management|"
r"penetrationstest|security[-\s]?incident|"
r"intrusion[-\s]?detection|firewall|"
r"tom|technisch[-\s]?organisatorische|"
r"iso[-\s]?2700[12]|bsi[-\s]?grundschutz|"
r"protokollier(ung)?|audit[-\s]?log|"
r"datensicherheit|netzwerksicherheit|"
r"patch[-\s]?management|update[-\s]?prozess|"
r"physische?\s+sicherheit|zutrittskontrolle)"
),
),
# DSE (DEUTLICH erweitert — wichtigster Bucket fuer aktuelle Audits)
(
"dse",
re.compile(
r"\b(art\.?\s*1[34]|"
r"datenschutzerklaerung|datenschutzhinweis|datenschutzinformation|"
r"informationspflicht|"
r"empfaenger(\s*oder\s*empfaengerkategorien)?|"
r"drittland.{0,30}(transfer|uebermittlung)|"
r"verantwortlich(er|en)\s+benennen|"
r"rechtsgrundlage.{0,30}(verarbeitung|nennen|angeben)|"
r"betroffenenrecht|"
r"art\.?\s*1[5-9]\s*dsgvo|art\.?\s*2[0-2]\s*dsgvo|"
r"art\.?\s*1[5-9]|art\.?\s*2[0-2]|"
r"auskunftsrecht|berichtigungsrecht|loeschungsrecht|"
r"einschraenkungsrecht|datenportabilitaet|widerspruchsrecht|"
r"einwilligung.{0,40}widerruf|"
r"datenschutzbeauftragt(er|en)?|dsb|"
r"aufsichtsbehoerde.{0,30}(benennen|nennen|kontakt)|"
r"beschwerderecht|art\.?\s*77|"
r"speicherdauer|loeschfrist|aufbewahrungsfrist|"
r"besondere\s+kategorien|art\.?\s*9\s*dsgvo|"
r"profiling|automatisierte\s+entscheidung|art\.?\s*22)"
),
),
# Process (erweitert)
(
"process",
re.compile(
r"\b(prozess|verfahren|workflow|"
r"durchfuehren|umsetzen|implementieren|"
r"schulung|mitarbeiterunterweis|sensibilisier|"
r"regelmaessig.{0,30}pruefen|"
r"kontinuierlich|laufend|fortlaufend|"
r"datenpannenmeldung|art\.?\s*3[34]|"
r"meldepflicht.{0,30}(behoerd|datenpannen|breach)|"
r"loeschkonzept|loeschroutine|"
r"hinweisgebersystem|whistleblow|hinschg|"
r"dsfa|datenschutz[-\s]?folgenabschaetz|art\.?\s*35|"
r"vvt|verzeichnis.{0,30}verarbeitung|art\.?\s*30|"
r"risiko(analyse|bewertung|management)|"
r"penetration.{0,30}durchfuehr|"
r"audit.{0,30}durchfuehr|"
r"kontrolle.{0,30}(durchfuehr|etablier)|"
r"nachweis(fuehrung|pflicht)|rechenschaft)"
),
),
]
def classify_v2(title: str, objective: str, tags: str | None = None) -> str:
text = " ".join(
s.lower() for s in (title or "", objective or "", tags or "") if s
)
for scope, pattern in RULES_V2:
if pattern.search(text):
return scope
return "other"
def main() -> int:
dsn = os.environ.get("DATABASE_URL")
if not dsn:
print("DATABASE_URL missing", file=sys.stderr)
return 1
conn = psycopg2.connect(dsn)
cur = conn.cursor()
# NUR den 'other'-Bucket aus v1 re-classifizieren
cur.execute("""
SELECT id, title, objective, tags
FROM compliance.canonical_controls
WHERE scope_doc_type = 'other'
AND merged_into_uuid IS NULL
""")
rows = cur.fetchall()
print(f"Re-classifying {len(rows):,} 'other'-bucket MCs with v2 heuristic...",
file=sys.stderr)
from collections import Counter
stats = Counter()
moved = 0
batch = []
for row_id, title, objective, tags in rows:
new_scope = classify_v2(title or "", objective or "", tags)
if new_scope != "other":
moved += 1
stats[new_scope] += 1
batch.append((new_scope, row_id))
if len(batch) >= 1000:
cur.executemany(
"UPDATE compliance.canonical_controls SET scope_doc_type=%s WHERE id=%s",
batch,
)
conn.commit()
print(f" re-classified {sum(stats.values()):,} so far, moved={moved:,}",
file=sys.stderr)
batch = []
if batch:
cur.executemany(
"UPDATE compliance.canonical_controls SET scope_doc_type=%s WHERE id=%s",
batch,
)
conn.commit()
print(f"\n=== v2 Re-classification: moved {moved:,} / {len(rows):,} from 'other' ===")
for scope, n in sorted(stats.items(), key=lambda x: -x[1]):
pct = 100 * n / max(1, len(rows))
print(f" {scope:25s} {n:>7,} ({pct:>4.1f}%)")
return 0
if __name__ == "__main__":
sys.exit(main())