fix(gateway): detect stale sys.modules via expected exports (#17648 follow-up)

[anuragbanerjee]

What does this PR do?

After hermes update, a gateway can keep running with old modules still in sys.modules while tools on disk import newer symbols. Users see import errors; session commands do not reload Python.

#18409 restarts the gateway when key source files look newer than at startup (mtime check). This change adds a second condition: if hermes_cli.config or the project utils package is already loaded but is missing cfg_get, atomic_replace, or base_url_host_matches, treat the process as stale and use the same restart path. It does not add broad try/except import fallbacks (#17935).

Related Issue

Refs #17648. Follow-up to #18409.

Source branch

Fork: github.com/anuragbanerjee/hermes-agent, branch fix/gateway-stale-sys-modules-exports. To check out locally: gh repo clone anuragbanerjee/hermes-agent -- -b fix/gateway-stale-sys-modules-exports (or add the fork as a remote and fetch that branch).

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• gateway/run.py: _loaded_modules_missing_expected_exports(); call from _detect_stale_code(); warn when boot mtime baseline is zero.
• tests/gateway/test_stale_code_self_check.py: tests for the export-based path.

How to Test

1. pytest tests/gateway/test_stale_code_self_check.py -v
2. Optional: long-lived gateway + update without process exit; next message should restart or succeed, not fail importing cfg_get.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform:

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

For New Skills

N/A

Screenshots / Logs

N/A

[teknium1]

Thanks for isolating the stale-module failure and adding focused coverage.

Problems

• Current main deliberately removed this exact per-message stale-code auto-restart mechanism in f2332d4f1 because it interrupted chats and dropped the triggering message. The current design instead records a boot fingerprint in gateway/run.py:20274-20279 and limits the guard to model switching in gateway/slash_commands.py:57-83.
• The proposed export list is incomplete: tests/test_stale_utils_module_import.py:52-63 reproduces the same failure when stale utils lacks env_float, which is not checked by the new helper. The linked June 24 report also identifies stale imports from tools.tool_backend_helpers and cron.jobs.

Suggested changes

• Re-scope any follow-up to current gateway/code_skew.py behavior without restoring automatic restarts that drop arbitrary inbound messages.
• Cover representative lazy-import consumers rather than treating three exports as a complete stale-module contract.

Automated hermes-sweeper review.

[teknium1]

This fixed list is not a complete stale-module contract: current regression coverage at tests/test_stale_utils_module_import.py:52-63 reproduces the same ImportError when stale utils lacks env_float, which this probe does not check. The linked June 24 report also identifies missing exports outside utils.

[anuragbanerjee]

Closing as superseded — the stale-code detection approach on main moved to git-revision fingerprinting (gateway/code_skew.py) instead of export-based sys.modules checks, and the auto-restart path this PR targeted was removed. The underlying gap (stale-module risk on regular agent turns, not just model switching) is worth tracking separately.