fix(sanitize): decode and sanitize bytes fields in sanitize_dict - #316

Summary

sanitize_dict sanitized only str values; bytes fields were decoded later by _json_default at serialization time — after the sanitization point. The same control byte was stripped from str fields but survived raw in bytes fields, then round-tripped back out through format_tool_result(...) → json.loads.

sanitize_dict now decodes bytes (UTF-8, replacing errors) at the sanitization point and sanitizes them as text, so both field types get identical treatment. _json_default stays as the last resort for bytes that reach json.dumps without passing through sanitize_dict.

Fixes #305

Before

from telegram_mcp.sanitize import sanitize_user_content, sanitize_dict

sanitize_user_content('\x1b[31mRED')   # '[31mRED'  (ESC stripped, as designed)
sanitize_dict({'b': b'\x1b[31mRED'})   # {'b': b'\x1b[31mRED'}  (ESC survived)

format_tool_result(...) then round-tripped the raw ESC sequence back out through json.loads — identical content got different treatment depending on the field type.

After

sanitize_dict({'b': b'\x1b[31mRED'})   # {'b': '[31mRED'}  (ESC stripped, same as str)
sanitize_dict({'b': b'caf\xc3\xa9\x00'})  # {'b': 'café'}  (decoded, control bytes stripped)

Bytes fields now return sanitized str — the same shape _json_default produced at serialization time, so the final JSON output is unchanged except that control bytes are gone.

Tests

• tests/test_sanitize.py::TestSanitizeDict::test_bytes_values_are_decoded_and_sanitized — ESC byte stripped from a bytes field, valid UTF-8 payload (café\x00) decoded and cleaned.
• Full suite: pytest -q → 1216 passed.

Risks

• sanitize_dict output type changes for bytes fields: bytes → sanitized str. This matches the final JSON serialization shape; no caller in the repo relies on bytes surviving sanitize_dict.

🤖 Generated with Claude Code