fix(agent): add encoding="utf-8" to os.fdopen() calls for cross-platform safety

[hashbender]

Problem

os.fdopen(fd, "w") without an explicit encoding parameter uses the platform default encoding. On Windows this is cp1252 (or mbcs), which silently corrupts non-ASCII characters — emoji, CJK, accented characters — when written to JSON files. The corrupted data is then read back with open(..., encoding="utf-8"), producing garbled output or UnicodeDecodeError.

This is the same class of cross-platform corruption bug as the write_text() without encoding pattern (PR NousResearch#54240).

Affected locations

1. agent/nous_rate_guard.py:120 — rate-limit state persistence via json.dump(state, f). The state dict can contain model names, provider slugs, or error messages with non-ASCII characters.

2. agent/shell_hooks.py:645 — shell hook metadata writes via json.dumps(data, ...). The data dict can contain user-provided hook names, command outputs, or environment variable values with Unicode.

Fix

Add encoding="utf-8" to both os.fdopen() calls, matching the pattern already used in hermes_cli/config.py (which passes **write_kw with encoding="utf-8").

Testing

• python3 -m py_compile agent/nous_rate_guard.py — OK
• python3 -m py_compile agent/shell_hooks.py — OK
• Both changes are single-line os.fdopen(fd, "w") → os.fdopen(fd, "w", encoding="utf-8")

---

Mirror-of: NousResearch#56033
NousResearch#56033

[tenki-reviewer]

Review Complete
No issues found!

Risk: 🟢 Low (5/100) — no findings · 4 LOC across 2 files

---

Clean, minimal fix: adds explicit encoding='utf-8' to two os.fdopen() calls to prevent locale-dependent encoding issues when writing JSON state files.

Files Reviewed (2 files)

agent/nous_rate_guard.py
agent/shell_hooks.py