fix(agent,gateway,cli,tools): add encoding="utf-8" to all .read_text() calls (47 instances)

[hashbender]

Problem

Path.read_text() without an explicit encoding parameter uses the platform default encoding. On Windows this is cp1252, which silently corrupts non-ASCII characters — emoji, CJK, accented characters — when reading JSON config files, auth tokens, session state, and other UTF-8 content.

This is the read-side counterpart to the write_text() without encoding bug (PR NousResearch#54240). Files written with encoding="utf-8" but read back without it produce garbled data on Windows.

Scope

47 instances across 18 files, all using the pattern:

# Before (uses platform default — cp1252 on Windows)
data = json.loads(path.read_text())

# After (explicit UTF-8)
data = json.loads(path.read_text(encoding="utf-8"))

Files changed:

Directory	File	Count
agent/	auxiliary_client.py	1
agent/	copilot_acp_client.py	1
agent/	shell_hooks.py	1
gateway/	run.py	14
gateway/	status.py	1
gateway/	dead_targets.py	1
tools/	skills_hub.py	6
tools/	skills_sync.py	1
tools/	managed_tool_gateway.py	1
tools/	xai_http.py	1
hermes_cli/	profiles.py	4
hermes_cli/	main.py	4
hermes_cli/	auth.py	3
hermes_cli/	doctor.py	3
hermes_cli/	uninstall.py	2
hermes_cli/	banner.py	1
hermes_cli/	service_manager.py	1
hermes_cli/	container_boot.py	1

Testing

• python3 -m py_compile on all 18 files — all OK
• Mechanical replacement: .read_text() → .read_text(encoding="utf-8")
• No behavioral change on Linux/macOS (UTF-8 is already the default)

---

Mirror-of: NousResearch#56115
NousResearch#56115

[tenki-reviewer]

Review Complete

Files Reviewed: 18
Findings: 1

By Severity:

• 🟠 High: 1

---

1 high-severity restart-loop regression and 3 medium encoding/consistency issues across 4 changed files. The removed _booted_from_restart guard reintroduces a denial-of-service path in the gateway.

Files Reviewed (18 files)

agent/auxiliary_client.py
agent/copilot_acp_client.py
agent/shell_hooks.py
gateway/dead_targets.py
gateway/run.py
gateway/status.py
hermes_cli/auth.py
hermes_cli/banner.py
hermes_cli/container_boot.py
hermes_cli/doctor.py
hermes_cli/main.py
hermes_cli/profiles.py
hermes_cli/service_manager.py
hermes_cli/uninstall.py
tools/managed_tool_gateway.py
tools/skills_hub.py
tools/skills_sync.py
tools/xai_http.py

[tenki-reviewer]

Risk: 🟠 High (72/100) — 1 high finding · 176 LOC across 18 files

---

Overview

This PR removes stale vendor integration code from gateway/run.py, hermes_cli/doctor.py, and related files, along with cross-platform encoding hardening. 4 issues were identified — 1 high, 3 medium.

High Severity

• Restart-loop regression in gateway/run.py:11433 — removal of the _booted_from_restart / _startup_time guard re-opens a path where the gateway can enter an infinite restart loop. Every restart triggers a redelivery attempt that may itself fail and re-trigger restart, creating a denial-of-service spiral.

Medium Severity

• Encoding mismatch in tools/skills_hub.py:1054 — write_text() without an explicit encoding parameter writes in locale-dependent encoding, while a sibling read_text(encoding='utf-8') assumes UTF-8. On non-UTF-8 Windows locales this produces garbled skill data.
• Same encoding mismatch in gateway/run.py:8223 — the .update_response file path.
• Orphaned test references in tests/gateway/test_restart_redelivery_dedup.py:250 — tests still assert on removed attributes _booted_from_restart and _startup_time, which will fail against the new code.

[tenki-reviewer]

🟠 Removed _booted_from_restart fallback reintroduces infinite /restart loop vulnerability (issue NousResearch#18528) (bug)

The PR removes three interconnected pieces of a belt-and-suspenders guard that prevented infinite restart loops when the .restart_last_processed.json dedup marker file goes missing:

1. Removed from __init__() (previously lines 2660-2670):

self._startup_time: float = time.time()
self._booted_from_restart: bool = False

2. Removed from start() (previously around line 6580):

if _restart_notification_pending() or planned_restart_notification_pending:
    self._booted_from_restart = True

3. Removed from _is_stale_restart_redelivery() (previously lines 11441-11446):
The 12-line fallback block that checked _booted_from_restart, _startup_time, and the 60-second post-boot window when the marker was missing — consuming the flag one-shot to suppress exactly one redelivered /restart per boot cycle.

The current code at line 11433-11434 unconditionally returns False when the marker doesn't exist:

if not marker_path.exists():
    return False

Impact: When the dedup marker is missing (e.g., atomic_json_write caught and logged an error at slash_commands.py:985-991, disk full, permission error, filesystem corruption) and Telegram re-delivers the /restart command (which happens when PTB's graceful-shutdown get_updates ACK fails), the gateway will process /restart again and immediately restart — creating the exact self-perpetuating infinite loop documented in issue NousResearch#18528.

Collateral: Three tests in tests/gateway/test_restart_redelivery_dedup.py (lines 250-317) still reference _booted_from_restart and _startup_time and were not updated. The test at line 250 (test_marker_missing_but_booted_from_restart_ignores_redelivery) will now fail because the suppression behavior it expects no longer exists.

💡 Suggestion: Restore the _booted_from_restart flag, _startup_time timestamp, and the fallback check in _is_stale_restart_redelivery(). Alternatively, if the guard is intentionally removed because a different mechanism now handles this (e.g., container restart backoff), document that clearly and remove/update the three tests in test_restart_redelivery_dedup.py.

📋 Prompt for AI Agents

In gateway/run.py, restore three removed pieces to prevent infinite /restart loops (issue NousResearch#18528):

1. In GatewayRunner.init() after line 2655, add: self._startup_time: float = time.time() and self._booted_from_restart: bool = False
2. In start() around line 6582, before await self._send_restart_notification(), add the flag-setting block that sets self._booted_from_restart = True when a restart notification is pending
3. In _is_stale_restart_redelivery() at line 11433, replace the unconditional return False (when marker is missing) with the belt-and-suspenders fallback that checks _booted_from_restart, _startup_time, and the 60-second window, consuming the flag one-shot
   Then update tests/gateway/test_restart_redelivery_dedup.py lines 250-317 to match the restored behavior.