fix(gateway): guard request_overrides None crash in message flow

[atongrun]

What does this PR do?

Fixes a gateway crash path where turn_route.get("request_overrides") can be None, but downstream agent code assumes a dict and calls .get("speed"). In messaging flows this breaks request construction before the provider call and can also trigger a secondary api_kwargs error while handling the failure.

Related Issue

Fixes #

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

• In gateway/run.py, normalize turn_route.get("request_overrides") to an empty dict before assigning it to the cached agent.
• In run_agent.py, guard the Anthropic fast_mode lookup with an empty-dict fallback.
• In run_agent.py, initialize api_kwargs = None before the retry loop and only dump API request debug data when it exists, preventing a secondary UnboundLocalError from masking the original failure.

How to Test

1. Configure a gateway flow where turn_route omits request_overrides or returns None.
2. Send a Telegram or gateway message that reaches GatewayRunner._run_agent().
3. Confirm the agent no longer crashes with NoneType object has no attribute get in _build_api_kwargs and does not emit the secondary UnboundLocalError for api_kwargs.
4. Manual validation performed on Ubuntu 24.04 against a live Telegram-backed Hermes gateway instance.

Checklist

Code

• I have read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this is not a duplicate
• My PR contains only changes related to this fix
• I have run pytest tests/ -q and all tests pass
• I have added tests for my changes
• I have tested on my platform: Ubuntu 24.04

Documentation & Housekeeping

• Documentation update not needed
• Config example update not needed
• CONTRIBUTING or AGENTS update not needed
• Cross-platform impact considered
• Tool descriptions or schemas update not needed

Screenshots / Logs

Observed failure before the fix in a live gateway session:

• AttributeError: NoneType object has no attribute get at run_agent.py while building API kwargs
• Followed by a secondary UnboundLocalError for api_kwargs during error handling

[teknium1]

Thanks for this careful bug report and fix, @atongrun! After reviewing main, all three crash paths this PR addresses are already resolved by prior commits.

Automated hermes-sweeper review — verifying against current main:

• fast_mode None-guard — run_agent.py:7660 already reads (self.request_overrides or {}).get("speed") == "fast" via commit 0bea60351 (PR fix: handle NoneType request_overrides in fast_mode check #7350, merged 2026-04-10).
• api_kwargs UnboundLocalError guard — run_agent.py:10080 already initializes api_kwargs = None before the retry loop via commit ea81aa2ee (PR fix: guard api_kwargs in except handler to prevent UnboundLocalError #7376, merged 2026-04-10).
• __init__ dict-normalization — run_agent.py:1135 already has self.request_overrides = dict(request_overrides or {}), introduced with the /fast feature in commit d416a6928.

The one gap (gateway's direct attribute assignment at gateway/run.py:10182 can still set request_overrides = None) is latent but harmless since the only consumer already uses (self.request_overrides or {}). Closing as implemented on main.