Pablan, as it stands

Self-hosted knowledge management for SMEs: a split-screen Markdown editor
whose sections an LLM refines while you write, and RAG question answering
over the documents that result. FastAPI + Postgres/pgvector on the back,
SvelteKit on the front, everything OpenAI-compatible and self-hostable.

Squashed into a single commit; the development history stays local.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CA43ZJda8Rbp2hKXNy8f6b

backend/tests/test_observability.py

@@ -0,0 +1,130 @@
+import json
+import logging
+
+import pytest
+from pydantic import BaseModel
+
+from app.config import get_settings
+from app.llm.client import chat_json
+from app.log import (
+    JsonFormatter,
+    apply_content_log_guard,
+    correlation_id,
+    safe_error,
+)
+from app.metrics import MetricsRegistry
+from tests.fake_openai import FakeOpenAI
+
+
+def test_metrics_registry_roundtrip() -> None:
+    registry = MetricsRegistry()
+    registry.inc("calls", {"role": "chat"})
+    registry.inc("calls", {"role": "chat"}, value=2)
+    registry.set_gauge("depth", 4.0)
+    registry.observe("seconds", 1.0, {"kind": "x"})
+    registry.observe("seconds", 3.0, {"kind": "x"})
+
+    snapshot = registry.snapshot()
+    assert snapshot["counters"]["calls"] == [{"labels": {"role": "chat"}, "value": 3.0}]
+    assert snapshot["gauges"]["depth"] == [{"labels": {}, "value": 4.0}]
+    hist = snapshot["histograms"]["seconds"][0]
+    assert hist == {
+        "labels": {"kind": "x"},
+        "count": 2,
+        "sum": 4.0,
+        "min": 1.0,
+        "max": 3.0,
+        "avg": 2.0,
+    }
+
+    registry.reset()
+    assert registry.snapshot() == {"counters": {}, "gauges": {}, "histograms": {}}
+
+
+def _format(record: logging.LogRecord) -> dict:
+    return json.loads(JsonFormatter().format(record))
+
+
+def test_json_formatter_includes_extras_and_correlation_id() -> None:
+    token = correlation_id.set("req-123")
+    try:
+        record = logging.LogRecord(
+            name="pablan.test",
+            level=logging.INFO,
+            pathname=__file__,
+            lineno=1,
+            msg="llm call",
+            args=(),
+            exc_info=None,
+        )
+        record.role = "chat"
+        record.duration_ms = 42
+        payload = _format(record)
+    finally:
+        correlation_id.reset(token)
+
+    assert payload["message"] == "llm call"
+    assert payload["level"] == "INFO"
+    assert payload["correlation_id"] == "req-123"
+    assert payload["role"] == "chat"
+    assert payload["duration_ms"] == 42
+    assert "ts" in payload
+
+
+def test_safe_error_strips_sql_parameters() -> None:
+    error = ValueError(
+        "insert failed [SQL: INSERT INTO messages ...] [parameters: ('secret content',)]"
+    )
+    sanitized = safe_error(error)
+    assert sanitized == "ValueError: insert failed"
+    assert "secret content" not in sanitized
+
+
+def test_safe_error_truncates() -> None:
+    sanitized = safe_error(RuntimeError("x" * 1000), limit=50)
+    assert len(sanitized) <= len("RuntimeError: ") + 50
+
+
+class Verdict(BaseModel):
+    done: bool
+
+
+async def test_llm_logs_contain_no_content(
+    fake_llm: FakeOpenAI, caplog: pytest.LogCaptureFixture
+) -> None:
+    """CLAUDE.md rule 12: with debug logging off (the default), neither the
+    prompt nor the model response may appear in any rendered log line."""
+    secret_prompt = "GEHEIM-PROMPT-77"
+    secret_response = '{"done": true, "leak": "GEHEIM-ANTWORT-88"}'
+    fake_llm.chat_responses.append({"content": secret_response})
+
+    # As in production: third-party SDK loggers are capped so they cannot
+    # dump request bodies even at global DEBUG level.
+    apply_content_log_guard()
+    with caplog.at_level(logging.DEBUG):
+        await chat_json([{"role": "user", "content": secret_prompt}], Verdict)
+
+    formatter = JsonFormatter()
+    rendered = "\n".join(formatter.format(record) for record in caplog.records)
+    assert "llm call" in rendered
+    assert "GEHEIM-PROMPT-77" not in rendered
+    assert "GEHEIM-ANTWORT-88" not in rendered
+
+
+async def test_debug_flag_enables_content_logging(
+    fake_llm: FakeOpenAI,
+    caplog: pytest.LogCaptureFixture,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """The documented never-in-production escape hatch actually works."""
+    monkeypatch.setenv("PABLAN_DEBUG_LOG_PROMPTS", "true")
+    get_settings.cache_clear()
+    try:
+        fake_llm.chat_responses.append({"content": '{"done": true}'})
+        with caplog.at_level(logging.DEBUG):
+            await chat_json([{"role": "user", "content": "SICHTBAR-99"}], Verdict)
+        formatter = JsonFormatter()
+        rendered = "\n".join(formatter.format(record) for record in caplog.records)
+        assert "SICHTBAR-99" in rendered
+    finally:
+        get_settings.cache_clear()