Phase 1 Step 4, file 2 of 18. Same cookbook as audit_routes (4a91814 +
883ef70) applied to banner_routes.py.
compliance/api/banner_routes.py (653 LOC) is decomposed into:
compliance/api/banner_routes.py (255) — thin handlers
compliance/services/banner_consent_service.py (298) — public SDK surface
compliance/services/banner_admin_service.py (238) — site/category/vendor CRUD
compliance/services/_banner_serializers.py ( 81) — ORM-to-dict helpers
shared between the
two services
compliance/schemas/banner.py ( 85) — Pydantic request models
Split rationale: the SDK-facing endpoints (consent CRUD, config
retrieval, export, stats) and the admin CRUD endpoints (sites +
categories + vendors) have distinct audiences and different auth stories,
and combined they would push the service file over the 500 hard cap.
Two focused services is cleaner than one ~540-line god class.
The shared ORM-to-dict helpers live in a private sibling module
(_banner_serializers) rather than a static method on either service, so
both services can import without a cycle.
Handlers follow the established pattern:
- Depends(get_consent_service) or Depends(get_admin_service)
- `with translate_domain_errors():` wrapping the service call
- Explicit return type annotations
- ~3-5 lines per handler
Services raise NotFoundError / ConflictError / ValidationError from
compliance.domain; no HTTPException in the service layer.
mypy.ini flips compliance.api.banner_routes from ignore_errors=True to
False, joining audit_routes in the strict scope. The services carry the
same scoped `# mypy: disable-error-code="arg-type,assignment"` header
used by the audit services for the ORM Column[T] issue.
Pydantic schemas moved to compliance.schemas.banner (mirroring the Step 3
schemas split). They were previously defined inline in banner_routes.py
and not referenced by anything outside it, so no backwards-compat shim
is needed.
Verified:
- 224/224 pytest (173 baseline + 26 audit integration + 25 banner
integration) pass
- tests/contracts/test_openapi_baseline.py green (360/484 unchanged)
- mypy compliance/ -> Success: no issues found in 123 source files
- All new files under the 300 soft target (largest: 298)
- banner_routes.py drops from 653 -> 255 LOC (below hard cap)
Hard-cap violations remaining: 16 (was 17).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 1 Step 4 follow-up addressing the debt flagged in the worked-example
commit (4a91814).
## mypy --strict policy
Adds backend-compliance/mypy.ini declaring the strict-mode scope:
Fully strict (enforced today):
- compliance/domain/
- compliance/schemas/
- compliance/api/_http_errors.py
- compliance/api/audit_routes.py (refactored in Step 4)
- compliance/services/audit_session_service.py
- compliance/services/audit_signoff_service.py
Loose (ignore_errors=True) with a migration path:
- compliance/db/* — SQLAlchemy 1.x Column[] vs
runtime T; unblocks Phase 1
until a Mapped[T] migration.
- compliance/api/<route>.py — each route file flips to
strict as its own Step 4
refactor lands.
- compliance/services/<legacy util> — 14 utility services
(llm_provider, pdf_extractor,
seeder, ...) that predate the
clean-arch refactor.
- compliance/tests/ — excluded (legacy placeholder
style). The new TestClient-
based integration suite is
type-annotated.
The two new service files carry a scoped `# mypy: disable-error-code="arg-type,assignment"`
header for the ORM Column[T] issue — same underlying SQLAlchemy limitation,
narrowly scoped rather than wholesale ignore_errors.
Flow: `cd backend-compliance && mypy compliance/` -> clean on 119 files.
CI yaml updated to use the config instead of ad-hoc package lists.
## Bugs fixed while enabling strict
mypy --strict surfaced two latent bugs in the pre-refactor code. Both
were invisible because the old `compliance/tests/test_audit_routes.py`
is a placeholder suite that asserts on request-data shape and never
calls the handlers:
- AuditSessionResponse.updated_at is a required field in the schema,
but the original handler didn't pass it. Fixed in
AuditSessionService._to_response.
- PaginationMeta requires has_next + has_prev. The original audit
checklist handler didn't compute them. Fixed in
AuditSignOffService.get_checklist.
Both are behavior-preserving at the HTTP level because the old code
would have raised Pydantic ValidationError at response serialization
had the endpoint actually been exercised.
## Integration test suite
Adds backend-compliance/tests/test_audit_routes_integration.py — 26
real TestClient tests against an in-memory sqlite backend (StaticPool).
Replaces the coverage gap left by the placeholder suite.
Covers:
- Session CRUD + lifecycle transitions (draft -> in_progress -> completed
-> archived), including the 409 paths for illegal transitions
- Checklist pagination, filtering, search
- Sign-off create / update / auto-start-session / count-flipping
- Sign-off 400 (invalid result), 404 (missing requirement), 409 (completed session)
- Get-signoff 404 / 200 round-trip
Uses a module-scoped schema fixture + per-test DELETE-sweep so the
suite runs in ~2.3s despite the ~50-table ORM surface.
Verified:
- 199/199 pytest (173 original + 26 new audit integration) pass
- tests/contracts/test_openapi_baseline.py green, OpenAPI 360/484 unchanged
- mypy compliance/ -> Success: no issues found in 119 source files
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 1 Step 4 of PHASE1_RUNBOOK.md, first worked example. Demonstrates
the router -> service delegation pattern for all 18 oversized route
files still above the 500 LOC hard cap.
compliance/api/audit_routes.py (637 LOC) is decomposed into:
compliance/api/audit_routes.py (198) — thin handlers
compliance/services/audit_session_service.py (259) — session lifecycle
compliance/services/audit_signoff_service.py (319) — checklist + sign-off
compliance/api/_http_errors.py ( 43) — reusable error translator
Handlers shrink to 3-6 lines each:
@router.post("/sessions", response_model=AuditSessionResponse)
async def create_audit_session(
request: CreateAuditSessionRequest,
service: AuditSessionService = Depends(get_audit_session_service),
):
with translate_domain_errors():
return service.create(request)
Services are HTTP-agnostic: they raise NotFoundError / ConflictError /
ValidationError from compliance.domain, and the route layer translates
those to HTTPException(404/409/400) via the translate_domain_errors()
context manager in compliance.api._http_errors. The error translator is
reusable by every future Step 4 refactor.
Services take a sqlalchemy Session in the constructor and are wired via
Depends factories (get_audit_session_service / get_audit_signoff_service).
No globals, no module-level state.
Behavior is byte-identical at the HTTP boundary:
- Same paths, methods, status codes, response models
- Same error messages (domain error __str__ preserved)
- Same auto-start-on-first-signoff, same statistics calculation,
same signature hash format, same PDF streaming response
Verified:
- 173/173 pytest compliance/tests/ tests/contracts/ pass
- OpenAPI 360 paths / 484 operations unchanged
- audit_routes.py under soft 300 target
- Both new service files under soft 300 / hard 500
Note: compliance/tests/test_audit_routes.py contains placeholder tests
that do not actually import or call the handler functions — they only
assert on request-data shape. Real behavioral coverage relies on the
contract test. A follow-up commit should add TestClient-based
integration tests for the audit endpoints. Flagged in PHASE1_RUNBOOK.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 1 Step 5 of PHASE1_RUNBOOK.md.
compliance/db/repository.py (1547 LOC) decomposed into seven sibling
per-aggregate repository modules:
regulation_repository.py (268) — Regulation + Requirement
control_repository.py (291) — Control + ControlMapping
evidence_repository.py (143)
risk_repository.py (148)
audit_export_repository.py (110)
service_module_repository.py (247)
audit_session_repository.py (478) — AuditSession + AuditSignOff
compliance/db/isms_repository.py (838 LOC) decomposed into two
sub-aggregate modules mirroring the models split:
isms_governance_repository.py (354) — Scope, Policy, Objective, SoA
isms_audit_repository.py (499) — Finding, CAPA, Review, Internal Audit,
Trail, Readiness
Both original files become thin re-export shims (37 and 25 LOC
respectively) so every existing import continues to work unchanged.
New code SHOULD import from the aggregate module directly.
All new sibling files under the 500-line hard cap; largest is
isms_audit_repository.py at 499 (on the edge; when Phase 1 Step 4
router->service extraction lands, the audit_session repo may split
further if growth exceeds 500).
Verified:
- 173/173 pytest compliance/tests/ tests/contracts/ pass
- OpenAPI 360 paths / 484 operations unchanged
- All repo files under 500 LOC
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 1 Step 3 of PHASE1_RUNBOOK.md. compliance/api/schemas.py is
decomposed into 16 per-domain Pydantic schema modules under
compliance/schemas/:
common.py ( 79) — 6 API enums + PaginationMeta
regulation.py ( 52)
requirement.py ( 80)
control.py (119) — Control + Mapping
evidence.py ( 66)
risk.py ( 79)
ai_system.py ( 63)
dashboard.py (195) — Dashboard, Export, Executive Dashboard
service_module.py (121)
bsi.py ( 58) — BSI + PDF extraction
audit_session.py (172)
report.py ( 53)
isms_governance.py (343) — Scope, Context, Policy, Objective, SoA
isms_audit.py (431) — Finding, CAPA, Review, Internal Audit, Readiness, Trail, ISO27001
vvt.py (168)
tom.py ( 71)
compliance/api/schemas.py becomes a 39-line re-export shim so existing
imports (from compliance.api.schemas import RegulationResponse) keep
working unchanged. New code should import from the domain module
directly (from compliance.schemas.regulation import RegulationResponse).
Deferred-from-sweep: all 28 class Config blocks in the original file
were converted to model_config = ConfigDict(...) during the split.
schemas.py-sourced PydanticDeprecatedSince20 warnings are now gone.
Cross-domain references handled via targeted imports (e.g. dashboard.py
imports EvidenceResponse from evidence, RiskResponse from risk). common
API enums + PaginationMeta are imported by every domain module.
Verified:
- 173/173 pytest compliance/tests/ tests/contracts/ pass
- OpenAPI 360 paths / 484 operations unchanged (contract test green)
- All new files under the 500-line hard cap (largest: isms_audit.py
at 431, isms_governance.py at 343, dashboard.py at 195)
- No file in compliance/schemas/ or compliance/api/schemas.py
exceeds the hard cap
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Squash of branch refactor/phase0-guardrails-and-models-split — 4 commits,
81 files, 173/173 pytest green, OpenAPI contract preserved (360 paths /
484 operations).
## Phase 0 — Architecture guardrails
Three defense-in-depth layers to keep the architecture rules enforced
regardless of who opens Claude Code in this repo:
1. .claude/settings.json PreToolUse hook on Write/Edit blocks any file
that would exceed the 500-line hard cap. Auto-loads in every Claude
session in this repo.
2. scripts/githooks/pre-commit (install via scripts/install-hooks.sh)
enforces the LOC cap locally, freezes migrations/ without
[migration-approved], and protects guardrail files without
[guardrail-change].
3. .gitea/workflows/ci.yaml gains loc-budget + guardrail-integrity +
sbom-scan (syft+grype) jobs, adds mypy --strict for the new Python
packages (compliance/{services,repositories,domain,schemas}), and
tsc --noEmit for admin-compliance + developer-portal.
Per-language conventions documented in AGENTS.python.md, AGENTS.go.md,
AGENTS.typescript.md at the repo root — layering, tooling, and explicit
"what you may NOT do" lists. Root CLAUDE.md is prepended with the six
non-negotiable rules. Each of the 10 services gets a README.md.
scripts/check-loc.sh enforces soft 300 / hard 500 and surfaces the
current baseline of 205 hard + 161 soft violations so Phases 1-4 can
drain it incrementally. CI gates only CHANGED files in PRs so the
legacy baseline does not block unrelated work.
## Deprecation sweep
47 files. Pydantic V1 regex= -> pattern= (2 sites), class Config ->
ConfigDict in source_policy_router.py (schemas.py intentionally skipped;
it is the Phase 1 Step 3 split target). datetime.utcnow() ->
datetime.now(timezone.utc) everywhere including SQLAlchemy default=
callables. All DB columns already declare timezone=True, so this is a
latent-bug fix at the Python side, not a schema change.
DeprecationWarning count dropped from 158 to 35.
## Phase 1 Step 1 — Contract test harness
tests/contracts/test_openapi_baseline.py diffs the live FastAPI /openapi.json
against tests/contracts/openapi.baseline.json on every test run. Fails on
removed paths, removed status codes, or new required request body fields.
Regenerate only via tests/contracts/regenerate_baseline.py after a
consumer-updated contract change. This is the safety harness for all
subsequent refactor commits.
## Phase 1 Step 2 — models.py split (1466 -> 85 LOC shim)
compliance/db/models.py is decomposed into seven sibling aggregate modules
following the existing repo pattern (dsr_models.py, vvt_models.py, ...):
regulation_models.py (134) — Regulation, Requirement
control_models.py (279) — Control, Mapping, Evidence, Risk
ai_system_models.py (141) — AISystem, AuditExport
service_module_models.py (176) — ServiceModule, ModuleRegulation, ModuleRisk
audit_session_models.py (177) — AuditSession, AuditSignOff
isms_governance_models.py (323) — ISMSScope, Context, Policy, Objective, SoA
isms_audit_models.py (468) — Finding, CAPA, MgmtReview, InternalAudit,
AuditTrail, Readiness
models.py becomes an 85-line re-export shim in dependency order so
existing imports continue to work unchanged. Schema is byte-identical:
__tablename__, column definitions, relationship strings, back_populates,
cascade directives all preserved.
All new sibling files are under the 500-line hard cap; largest is
isms_audit_models.py at 468. No file in compliance/db/ now exceeds
the hard cap.
## Phase 1 Step 3 — infrastructure only
backend-compliance/compliance/{schemas,domain,repositories}/ packages
are created as landing zones with docstrings. compliance/domain/
exports DomainError / NotFoundError / ConflictError / ValidationError /
PermissionError — the base classes services will use to raise
domain-level errors instead of HTTPException.
PHASE1_RUNBOOK.md at backend-compliance/PHASE1_RUNBOOK.md documents
the nine-step execution plan for Phase 1: snapshot baseline,
characterization tests, split models.py (this commit), split schemas.py
(next), extract services, extract repositories, mypy --strict, coverage.
## Verification
backend-compliance/.venv-phase1: uv python install 3.12 + pip -r requirements.txt
PYTHONPATH=. pytest compliance/tests/ tests/contracts/
-> 173 passed, 0 failed, 35 warnings, OpenAPI 360/484 unchanged
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Containers are on multiple networks (breakpilot-network, coolify,
gokocgws...). Without traefik.docker.network, Traefik randomly picks
a network and may choose breakpilot-network where it has no access.
This label forces Traefik to always use the coolify network.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Traefik routes traffic via the 'coolify' bridge network, so services
that need public domain access must be on both breakpilot-network
(for inter-service communication) and coolify (for Traefik routing).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
SQLAlchemy 2.x requires raw SQL strings to be explicitly wrapped
in text(). Fixed 16 instances across 5 route files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Switch to ${COMPLIANCE_DATABASE_URL} for admin-compliance, backend, SDK, crawler
- Add DATABASE_URL to admin-compliance environment
- Switch ai-compliance-sdk from QDRANT_HOST/PORT to QDRANT_URL + QDRANT_API_KEY
- Add MINIO_SECURE to compliance-tts-service
- Update .env.coolify.example with new variable patterns
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace bp-core-postgres with POSTGRES_HOST env var
- Replace bp-core-qdrant with QDRANT_HOST env var
- Replace bp-core-minio with S3_ENDPOINT/S3_ACCESS_KEY/S3_SECRET_KEY
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add docker-compose.coolify.yml (8 services), .env.coolify.example,
and Gitea Action workflow for Coolify API deployment. Removes
core-health-check and docs. Adds Traefik labels for
*.breakpilot.ai domain routing with Let's Encrypt SSL.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replaced bare imports with safe_import_router pattern — if one sub-router
fails to import (e.g. missing dependency), other routers still load.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
from __future__ import annotations breaks Pydantic BaseModel runtime type
evaluation. Replaced str | None → Optional[str], list[str] → List[str] etc.
in control_generator.py, anchor_finder.py, control_generator_routes.py.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds migration_runner.py that executes pending migrations from
migrations/ directory when backend-compliance starts. Tracks applied
migrations in _migration_history table.
Handles existing databases: detects if tables from migrations 001-045
already exist and seeds the history table accordingly, so only new
migrations (046+) are applied.
Skippable via SKIP_MIGRATIONS=true env var.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Migration 045: Seed 10 controls (AUTH, NET, SUP, LOG, WEB, DATA, CRYP, REL)
with 39 open-source anchors into the database
- Backend: POST/PUT/DELETE endpoints for canonical controls CRUD
- Frontend proxy: PUT and DELETE methods added to canonical route
- Frontend: Control Library with create/edit/delete UI, full form with
open anchor management, scope, requirements, evidence, test procedures
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The backend mounts the compliance router at /api/compliance, so canonical
control endpoints are at /api/compliance/v1/canonical/*, not /api/v1/canonical/*.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add phase_security() with 15 documents across 3 sub-phases:
- J1: 7 NIST standards (SP 800-53, 800-218, 800-63, 800-207, 8259A/B, AI RMF)
- J2: 6 OWASP projects (Top 10, API Security, ASVS, MASVS, SAMM, Mobile Top 10)
- J3: 2 ENISA guides (Procurement Hospitals, Cloud Security SMEs)
All documents are commercially licensed (Public Domain / CC BY / CC BY-SA).
Wire up 'security' phase in dispatcher and workflow yaml.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
download_pdf() and extract_gesetz_html() now return 0 on failure and clean up
partial files. This prevents set -euo pipefail from aborting the entire script
when a single download fails (e.g. EUR-Lex timeout, BSI redirect).
Root cause of H2 EU loop only processing 1 document in Run #724: first failed
download_pdf returned 1, triggering set -e script abort.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
docker cp fails when target dir doesn't exist in a created container.
Copy scripts to /workspace_scripts, then cp them at container start.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The runner container can't access host paths directly, so the
deploy dir scripts were always stale. Now uses docker create +
docker cp + docker start to copy the freshly checked-out scripts
into the ingestion container before starting it.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The RAG workflow mounts scripts from /opt/breakpilot-compliance/scripts
(deploy dir) but this may not have the latest fixes if CI hasn't
deployed yet. Add explicit git pull before running ingestion.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- collection_count() returns 0 (not ?) on failure — fixes arithmetic error
- Pass QDRANT_API_KEY to ingestion container for dedup checks
- Include api-key header in collection_count() and dedup scroll queries
- Lower large-file threshold to 256KB (EGBGB 310KB was timing out)
- More targeted EGBGB XML extraction (Art. 246a + Anlage only)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Critical bug fix: mandatoryDocuments in Hard-Trigger-Rules used UPPERCASE
names (VVT, TOM, DSE) that never matched lowercase ScopeDocumentType keys
(vvt, tom, dsi). This meant no trigger documents were ever recognized as
mandatory in buildDocumentScope().
- Add normalizeDocType() mapping function with alias support
(DSE→dsi, LOESCHKONZEPT→lf, DSR_PROZESS→betroffenenrechte, etc.)
- Fix buildDocumentScope() to use normalized doc types
- Fix estimateEffort() to use lowercase keys matching ScopeDocumentType
- Add 2 tests for UPPERCASE normalization and alias resolution
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Extended timeout (15 min) for files > 500KB (BGB is 1.5MB)
- upload_file returns 0 even on failure so set -e doesn't kill script
- Failed uploads are still counted and reported in summary
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The gesetze phase failed because it expects text files created by the
download phase. Now the workflow automatically runs download first for
any phase that depends on it. Also adds git and python3 to the alpine
container for repo cloning and text extraction.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- admin-compliance/Dockerfile: mkdir -p public before build
- developer-portal/Dockerfile: mkdir -p public before build
(fixes "failed to calculate checksum /app/public: not found")
- docker-compose.hetzner.yml: Override core-health-check to exit
immediately (Core doesn't run on Hetzner)
- Network override: external:false (auto-create breakpilot-network)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The runner container has Docker socket but no host filesystem access.
docker compose needs to read YAML files, so run build+deploy inside
a helper container that has both Docker socket and the deploy dir mounted.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Problems fixed:
1. Deploy step couldn't access /opt/breakpilot-compliance (host path not
mounted in runner container). Now uses alpine/git helper container with
host bind-mount for git ops, then docker compose with host paths.
2. breakpilot-network was external:true but Core doesn't run on Hetzner.
Override in hetzner.yml creates the network automatically.
3. core-health-check blocks startup waiting for Core. Override in
hetzner.yml makes it exit immediately.
4. RAG ingestion script now respects RAG_URL/QDRANT_URL env vars.
5. RAG workflow discovers network dynamically from running containers.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Temporary commit to discover Docker container names and networks
on Hetzner, since breakpilot-network doesn't exist there.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Instead of trying to connect the runner to breakpilot-network,
spawn a new alpine container directly on it via docker run.
Added debug output for network/container visibility.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Join breakpilot-network so bp-core-rag-service is reachable
- Make RAG_URL/QDRANT_URL in script respect env vars (${VAR:-default})
- Remove complex fallback logic — fail fast if network not available
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The runner container doesn't always have /opt/breakpilot-compliance mounted.
Use the git-cloned workspace (current dir) and add multi-fallback for RAG API
URL (container network → localhost → host.docker.internal).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Phase H now includes:
- 16 German laws (PAngV, VSBG, ProdHaftG, BDSG, HGB, AO, DDG, TKG, etc.)
- 15 EUR-Lex EU laws (DSGVO, Consumer Rights Dir, Sale of Goods Dir,
E-Commerce Dir, Unfair Terms Dir, DMA, NIS2, Product Liability Dir, etc.)
- 2 NIST frameworks (CSF 2.0, Privacy Framework 1.0)
- 1 HLEG Ethics Guidelines
Updated rag-sources.md with complete inventory of already-ingested vs
new documents, plus Layer 2-5 TODO roadmap.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Runner needs access to /opt/breakpilot-compliance and Docker network
for RAG service (bp-core-rag-service:8097). Falls back to
host.docker.internal if container network unavailable.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
