In `@infrahub_sdk/file_handler.py`:
- Around line 98-118: The handler handle_error_response should guard against
non-JSON responses by wrapping each exc.response.json() call in a try/except
(catching json.JSONDecodeError or ValueError) and falling back to
exc.response.text; for 401/403, if json parsing fails treat the body text as a
single message (or empty) when building the AuthenticationError message, and for
404 use the fallback text as the detail when constructing NodeNotFoundError;
preserve the original HTTPStatusError as the cause (raise ... from exc).

In `@infrahub_sdk/node/node.py`:
- Around line 1160-1163: The error raised when self._file_object_support is true
and self._file_content is None references the wrong helper method name; update
the ValueError text in the block that checks self._file_object_support /
self._file_content (near usage of self._schema.kind) to instruct callers to use
the correct methods select_file_for_upload() or select_content_for_upload()
instead of set_file().
- Around line 2049-2052: The error message inside the check for
self._file_object_support and self._file_content should reference the actual
synchronous setter method name instead of the non-existent set_file(); update
the message to instruct callers to use the real sync setter (e.g.,
set_file_content()) — modify the ValueError string that currently mentions
set_file() so it references set_file_content() (or the actual sync method used
in this class) and keep the rest of the message intact; target the block that
checks _file_object_support and _file_content and update the message formatting
around _schema.kind accordingly.

In `@infrahub_sdk/protocols.py`:
- Around line 110-115: The CoreFileObject class (and the other affected
generated blocks referenced at lines 234-237, 675-680, 799-802) were edited
manually but this file is generated; revert any hand edits to these definitions
and re-run the generation task or update the source schema so the generator
emits the correct definitions (e.g., restore CoreFileObject and the other
generated classes/fields to match the generator output), do not make direct
changes in infrahub_sdk/protocols.py, and commit the regenerated file instead.

In `@infrahub_sdk/client.py`:
- Around line 997-1033: Summary: The method _execute_graphql_with_file mutates
the caller's variables dict by assigning variables["file"] = None; make a copy
first to avoid leaking state. Fix: inside _execute_graphql_with_file (and the
other similar block) replace the in-place modification with a shallow copy
(e.g., vars_copy = dict(variables) or variables.copy() after handling None) and
set vars_copy["file"] = None, then use vars_copy for the request; ensure you
handle variables being None before copying. Reference symbols:
_execute_graphql_with_file, variables, "file".
- Around line 1060-1090: The multipart upload retries can send an empty file
because the passed file-like object has been consumed; in the async
_post_multipart method (decorated with handle_relogin) seek the incoming
file_content to the start (file_content.seek(0)) before calling
MultipartBuilder.build_payload so the payload is built from the beginning of the
stream, and apply the identical change to the sync counterpart (the sync
_post_multipart or similarly named method used elsewhere) and any other
occurrences (the other multipart methods around the second block referenced) so
retries re-upload the full file.

In `@infrahub_sdk/ctl/branch.py`:
- Around line 122-123: The two occurrences of "pc.created_by" (used in
proposed_change_table.add_row and format_timestamp) currently use a blanket "#
type: ignore[attr-defined]" without justification; fix this by either updating
the type definitions for the object that pc is (so created_by is declared on
that model) or by replacing the ignore with an explicit, checkable
assertion/cast (e.g., assert isinstance(pc, ExpectedProposedChange) or cast to
the model type) and add a short inline comment explaining why created_by is
guaranteed to exist (e.g., "created_by is populated by the API vX response for
created proposed changes"); update the annotation for pc or add the assert/cast
next to the accesses of pc.created_by so mypy no longer needs the ignore and
future readers see the rationale.

In `@infrahub_sdk/node/node.py`:
- Around line 249-275: The docstring example for upload_from_bytes is unsafe
because it stores a file handle in self._file_content and then closes it (via
the with block) before node.save() runs, causing I/O errors; update the example
and text in upload_from_bytes to show one of two correct patterns: (a) read the
file into bytes and call upload_from_bytes(content=f.read(), name=...) outside a
with, or (b) keep the handle open and call node.save() inside the with block
(e.g., with open(..., "rb") as f: node.upload_from_bytes(content=f, name="...");
node.save()). Mention that this applies to nodes inheriting from CoreFileObject
and reference upload_from_bytes, self._file_content and save() in the docstring.

In `@tests/unit/sdk/test_file_object.py`:
- Around line 78-105: The test uses type: ignore[union-attr] on attributes like
node.contract_start and node.contract_end in
test_node_create_with_file_uses_multipart (and the other listed tests) without
justification; replace each bare ignore by either asserting/casting the node to
the correct runtime type or adding a brief justification comment: e.g., use an
explicit runtime assertion (assert isinstance(node, InfrahubNode) or cast via
typing.cast) before accessing .contract_start/.contract_end, or add a one-line
comment explaining why the union attribute is present and safe (e.g., "# safe:
test constructs node with schema that guarantees contract_start attribute");
update the occurrences around InfrahubNode/InfrahubNodeSync instantiation so
mypy no longer needs type: ignore or the ignore is accompanied by the short
justification comment.