Context

OpenRouter announced beta support for response caching:
https://openrouter.ai/announcements/response-caching

This lets clients opt into caching completed responses for identical OpenRouter requests. Hermes already sends OpenRouter attribution headers; this PR adds a small, explicit opt-in path for the new cache headers without changing default behaviour.

The intent is to make response caching available for workflows where repeat requests are expected and safe, such as retries, test runs, cron/batch jobs, and stable auxiliary tasks. It stays disabled by default because live agent conversations and fresh tool-context workflows should not silently change transport behaviour.

Summary

• add a shared OpenRouter header helper that preserves existing attribution headers
• add optional OpenRouter response caching headers behind explicit config/env opt-in
• apply the helper to the main agent OpenRouter client and auxiliary OpenRouter clients

Config

Default remains disabled:

openrouter:
  response_cache:
    enabled: false
    ttl_seconds: null

To enable:

openrouter:
  response_cache:
    enabled: true
    ttl_seconds: 600

Environment overrides:

HERMES_OPENROUTER_RESPONSE_CACHE=true
HERMES_OPENROUTER_RESPONSE_CACHE_TTL=600

Tests

• python -m pytest tests/agent/test_openrouter_headers.py tests/agent/test_auxiliary_client.py tests/run_agent/test_run_agent.py -q
• python -m pytest tests/hermes_cli/test_config.py tests/hermes_cli/test_config_validation.py tests/hermes_cli/test_config_drift.py -q
• python -m py_compile agent/openrouter_headers.py agent/auxiliary_client.py run_agent.py hermes_cli/config.py
• git diff --check