fix(cv-import): bound source retention and guard worker recovery

tools/summarizer/Dockerfile.cv-import-dev

@@ -0,0 +1,18 @@
+# Isolated development image only. Supply an explicitly inspected existing runtime image ID.
+ARG CV_IMPORT_BASE_IMAGE=jobtracker-cv-import-legacy:required
+FROM ${CV_IMPORT_BASE_IMAGE}
+USER root
+COPY requirements-cv-import.lock /tmp/cv-import.lock
+COPY wheels/ /tmp/cv-import-wheels/
+RUN python -m venv /opt/cv-import && /opt/cv-import/bin/python -m pip install --no-index --find-links=/tmp/cv-import-wheels --require-hashes -r /tmp/cv-import.lock
+COPY app.py parser_child.py cv_import_ai.py /app/
+ENV CV_IMPORT_PYTHON=/opt/cv-import/bin/python \
+    CV_IMPORT_AI_ENABLED=false \
+    SKIP_MODEL_LOAD=1 \
+    AI_SERVICE_SKIP_MODEL_LOAD=1 \
+    HF_HUB_OFFLINE=1 \
+    TRANSFORMERS_OFFLINE=1 \
+    PYTHONDONTWRITEBYTECODE=1
+WORKDIR /app
+USER 10001
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8001"]

tools/summarizer/app.py

@@ -1222,6 +1222,8 @@ def _parser_environment(temp_root: str) -> dict[str, str]:
         "TMPDIR": temp_root,
         "TEMP": temp_root,
         "TMP": temp_root,
+        "HOME": temp_root,
+        "USERPROFILE": temp_root,
     })
     return environment
 

tools/summarizer/cv_import_ai.py

@@ -5,7 +5,7 @@ import json
 import os
 import re
 from urllib.parse import urlsplit
-from urllib.request import Request, urlopen
+from urllib.request import Request, build_opener, HTTPRedirectHandler, ProxyHandler
 
 
 @dataclass(frozen=True)
@@ -51,10 +51,15 @@ def _validate_config(config):
             raise ValueError("Only local provider addresses are supported.")
 
 
+class _NoRedirect(HTTPRedirectHandler):
+    def redirect_request(self, req, fp, code, msg, headers, newurl):
+        raise ValueError("Local inference redirects are not allowed.")
+
+
 def _ollama(config, payload):
     request = Request(config.base_url.rstrip("/") + "/api/generate", data=json.dumps(payload).encode(), headers={"Content-Type": "application/json"})
     # The application adapter never installs/pulls models or changes a model cache.
-    with urlopen(request, timeout=config.timeout_seconds) as response:
+    with build_opener(ProxyHandler({}), _NoRedirect()).open(request, timeout=config.timeout_seconds) as response:
         body = response.read(256 * 1024 + 1)
     if len(body) > 256 * 1024:
         raise ValueError("Model output exceeded its bound.")

tools/summarizer/parser_child.py

@@ -377,6 +377,9 @@ def main() -> int:
         data = Path(args.input).read_bytes()
         if args.mode == "document":
             os.environ.update({"HF_HUB_OFFLINE": "1", "HF_HUB_DISABLE_TELEMETRY": "1", "TRANSFORMERS_OFFLINE": "1"})
+            # Cache trusted platform metadata before the document sandbox denies Linux uname.
+            import platform
+            platform.platform()
             sys.addaudithook(_deny_import_side_effects)
             result = extract_document(args.extension.lower(), data)
         else:

tools/summarizer/tests/test_app.py

@@ -924,3 +924,10 @@ def test_endpoints_stay_open_when_no_token_is_configured(monkeypatch):
     response = client.post("/summarize", json={"text": "Platform engineering role."})
 
     assert response.status_code == 503
+
+
+def test_parser_home_is_private_work_directory_not_host_model_cache(monkeypatch):
+    module = load_app_module(monkeypatch)
+    environment = module._parser_environment("private-parser-work")
+    assert environment.get("HOME") == "private-parser-work"
+    assert environment.get("USERPROFILE") == "private-parser-work"