oh-my-hermes: An operational layer adding model tuning, parallel execution, and memory hierarchy to Hermes Agent

It adds a specialized operational layer on top of Hermes Agent's natural language interface. It provides routing capabilities that analyze requests to classify task types and automatically assign appropriate models and workflows. It does not replace Hermes's existing native skills, but instead establishes clear evidence boundaries to transparently manage execution records.

For coding tasks, it supports model-specific prompt tuning and parallel execution. It applies calibration blocks tailored to 13 model families and separates task units without file conflicts to process them simultaneously in different worktrees. It reports achieving identical results with the same GPT-6 Astra model, reducing costs by 84% and time by 78%.

108 omh-* expert skills are automatically loaded at runtime. When specific domain requests such as frontend, backend, Rust, or security reviews are detected, the corresponding skills are activated to perform quality gates and verification procedures. Even when delegating external executors like Codex or Claude Code via Maestro, the status and cost of each lane are tracked in real time.

It provides an integrated long-term memory system and optimized workflow packages. Users can directly define and override models and effort levels for 9 task categories, including ultrabrain, deep, and architect. It simplifies complex agent workflows while clearly distinguishing the verification status of each step to ensure reliability.