hermes - 💡(How to fix) Fix Roadmap: add evals and behavioral regression harness

Summary

Add an evaluation and regression harness for agent behavior, tool schemas, slash commands, plugins, and provider interactions.

Proposed scope

• Add golden transcript tests for CLI, gateway, and tool-call flows.
• Snapshot tool schemas and slash command registry outputs.
• Simulate agent-loop behavior with fake providers.
• Add plugin contract tests.
• Add evaluations for expected tool selection and call sequencing.
• Track cost and latency regressions where test infrastructure can do so deterministically.

Acceptance criteria

• Core agent and command behavior can be regression-tested without real model calls.
• Tool schema and command registry drift is caught in CI.
• Contributors have a clear place to add behavioral regression coverage.