hermes - 💡(How to fix) Fix [Feature] Allow autonomous coding delegation skills to consume central .providers config

[Feature] Allow autonomous coding delegation skills to consume central .providers config

Currently, the autonomous coding delegation skills (openhands, claude-code, codex, opencode, etc.) completely bypass Hermes' provider system.

When delegating work, you are forced to manually construct environment variables every time:

LLM_MODEL=deepseek/deepseek-v4-flash \
LLM_API_KEY=*** \
openhands --headless --json --override-with-envs ...

This is true even when the user has already defined the provider cleanly in ~/.hermes/config.yaml:

providers:
  deepseek:
    provider: deepseek
    model: deepseek-v4-flash
    api_key: ${DEEPSEEK_API_KEY}

Requested improvement

These skills should be able to reference Hermes providers instead of requiring raw LLM_MODEL / LLM_API_KEY construction.

Possible approaches:

• Support --model <provider-name> or --provider <name> when invoking the skill (e.g. hermes -s openhands --model deepseek)
• Have the skill documentation/examples show how to pull from the central config
• Add a lightweight helper in the skill that can resolve a Hermes provider name into the correct env vars for the external CLI
• Longer term: a more integrated delegation path that doesn't require shelling out with manual env var injection

This would remove duplication and make using custom / non-mainstream providers (native DeepSeek, local models, self-hosted endpoints, etc.) much more consistent with the rest of Hermes.

Related context:

• These skills were added as "model-agnostic" options (see #477 for OpenHands)
• They are currently implemented as usage recipes that shell out via the terminal tool