In `@server/server.py`:
- Around line 113-117: The current broad "except Exception as exc" block
re-raises an HTTPException without chaining, which loses the original traceback;
modify the handler to either catch specific agent/SDK exceptions (e.g.,
AgentExecutionError or SDK-specific exceptions) or, if you must catch Exception,
re-raise the HTTPException using exception chaining (raise
HTTPException(status_code=500, detail=f"Agent execution failed: {exc}") from
exc) so the original traceback is preserved; update the except block that
contains "except Exception as exc" and reference HTTPException to implement this
change.
- Around line 28-30: The current open("config.json") call loads CONFIGS from a
path relative to the CWD which breaks when the server is started from the repo
root; change the file lookup in server.py to resolve the config.json path
relative to the module file (use __file__ or
pathlib.Path(__file__).resolve().parent) before opening so CONFIGS and
config_json_str always read the module-local config.json regardless of working
directory.

In `@deploy/deployment.yaml`:
- Around line 21-55: Add a hardened securityContext to the Deployment's
container (name: oape-server) by setting runAsNonRoot: true, runAsUser to a
non-root UID, and enforcing readOnlyRootFilesystem: true while dropping all
capabilities; then provide a writable /tmp by adding a volumeMount for
mountPath: /tmp (readOnly: false) using an emptyDir volume (e.g., name: tmp-dir)
and add the corresponding volumes entry with emptyDir in the Pod spec so the
container has a writable temp directory despite a read-only root filesystem.

In `@server/server.py`:
- Around line 5-6: The usage snippet references the wrong module path; replace
"uvicorn api.server:app --reload" with the correct module path "uvicorn
server.server:app --reload" (or update to whichever importable module exposes
the FastAPI app) and ensure any other docs or comments that mention "api.server"
are updated to "server.server" so the uvicorn command imports the app correctly.
- Around line 52-58: CONVERSATION_LOG is hard-coded to /tmp which is unsafe;
make the log location configurable (e.g., read from an env var or config value)
and ensure the directory is created with private permissions before creating the
FileHandler. Update the code that defines CONVERSATION_LOG to derive the Path
from a config/env (falling back to a safe default inside the application data
dir), create the parent directory with mode 0o700 (os.makedirs(path,
exist_ok=True) + Path.chmod) and only then instantiate conv_logger, _handler and
add the FileHandler so the file is written in a non-world-writable, private
directory. Ensure conv_logger and _handler continue to be used as before.
- Around line 121-146: The messages_log entries can include objects (e.g.,
block.__dict__, message.__dict__) that are not JSON-serializable; update the
AssistantMessage and fallback branches (the block handling and the final else
handling) and the ResultMessage entry creation to sanitize content before
appending to messages_log. Implement or call a small sanitizer (e.g.,
sanitize_for_json) that recursively converts non-serializable objects to
primitives/strings (or uses json.dumps+json.loads with default=str) and apply it
to block_dict, msg_dict, and message.result (and any other content fields) so
every entry appended to messages_log is JSON-serializable; keep conv_logger
logging unchanged except use the sanitized representation for JSON output.

In `@server/server.py`:
- Line 98: The agent is hard-coded with permission_mode="bypassPermissions"
which disables permission checks; change this to read from a configurable source
(e.g., an environment variable like PERMISSION_MODE or a config setting) and
validate allowed values before passing to the agent initialization (replace the
literal "bypassPermissions" used where permission_mode is set). Default to a
restrictive mode (e.g., "enforcePermissions") when the env/config is absent,
emit a startup warning if a permissive mode like "bypassPermissions" is
selected, and add brief documentation/comments about the security implications
so production deployments use the safer default.
- Line 47: Remove the stray debug print by deleting the print(PLUGIN_DIR) call
in server.py (it runs on every import); if you need that info, replace it with a
logger.debug call instead (use the module logger or obtain one via
logging.getLogger(__name__)) so it no longer writes to stdout at import time.