fix(hindsight): specify UTF-8 encoding for file I/O on Windows

[jeeaay]

Problem

On Windows with CJK system locales (e.g. Chinese GBK), pathlib.Path.read_text() defaults to the system encoding instead of UTF-8. This causes a UnicodeDecodeError when the hindsight memory provider reads .env or JSON config files containing non-ASCII characters.

UnicodeDecodeError: 'gbk' codec can't decode byte 0x94 in position 661: illegal multibyte sequence

Fix

Explicitly pass encoding="utf-8" to all 3 read_text() and 1 write_text() calls in plugins/memory/hindsight/__init__.py that handle .env and JSON config files.

This is consistent with the existing write_text(..., encoding="utf-8") call already present at line 556 in the same file.

Affected lines

Line	Change
749	config_path.read_text() → read_text(encoding="utf-8")
898	env_path.read_text() → read_text(encoding="utf-8")
928	env_path.read_text() → read_text(encoding="utf-8")
941	env_path.write_text(...) → write_text(..., encoding="utf-8")

Environment

• OS: Windows 10, Chinese (GBK) locale
• Python: 3.11.15
• Hermes: v2026.7.1

[teknium1]

Thanks for the targeted Windows compatibility fix. The premise is confirmed on current main: HindsightMemoryProvider.save_config still reads JSON without an encoding at plugins/memory/hindsight/__init__.py:749, and post_setup still has default-encoding .env reads/writes at lines 898, 928, and 941. This matches the four lines in commit d0dbdd807a3a.

Problems

• The PR does not add regression coverage. Existing setup coverage is in tests/plugins/memory/test_hindsight_provider.py:503-625, but no test exercises UTF-8 non-ASCII JSON/.env content across the changed paths.

Suggested changes

• Add a focused temporary-HERMES_HOME regression covering save_config and local-embedded setup with UTF-8 non-ASCII fixtures, ensuring these paths continue to specify UTF-8.

This is an automated hermes-sweeper review.

[teknium1]

Merged via PR #71078 — your commit(s) were cherry-picked onto current main with your authorship preserved in git log (rebase merge). This PR was part of the class-wide close-out of bare read_text/write_text calls: all 139 remaining sites now pass explicit encoding, and a new CI linter rule prevents regressions. Thanks for the contribution!