hermes - 💡(How to fix) Fix Skill triggering is purely prompt-driven with no code-level enforcement

Problem

Skill triggering in Hermes is entirely dependent on the model's voluntary decision. The system prompt includes <available_skills> with descriptions, and the model is instructed to load relevant skills via skill_view, but there is no code-level enforcement — the model can ignore this instruction and answer directly.

This means:

1. Skills are effectively optional. A perfectly authored skill can be completely bypassed if the model decides not to call skill_view.
2. No user control. Users cannot force the model to follow a specific skill workflow.
3. Skill author intent is unenforceable. Detailed workflows written for regulated or complex processes have no guarantees of execution.
4. Both TUI and WebUI lack enforcement. Neither frontend provides a mechanism to manually attach a skill before a turn.

Evidence

• Skills are listed in the system prompt as text (the <available_skills> block)
• The model receives the skill_view tool like any other tool
• If the model chooses not to call skill_view, the skill content never loads
• There is no pre-processing layer that intercepts user messages and enforces skill matching

Questions / Requests

1. Is there an architectural reason skills are purely prompt-driven?
2. Could a rule-based matching layer be added that intercepts user input before the model sees it, and automatically loads relevant skills based on keyword/pattern matching?
3. Could a /use <skill> command be added that forces a specific skill into the context for the next turn?

This is affecting both the TUI and WebUI frontends, and both have been notified separately (fathah/hermes-desktop#389, nesquena/hermes-webui#2977). But the root cause appears to be in the core agent architecture.

Related

• fathah/hermes-desktop#389
• nesquena/hermes-webui#2977