Pro · agent skills, from real work
Skills from real work, not evals.
Every skill here was developed over months of real, day-to-day work in Hermes Agent — the disciplines, pitfalls, and hard-won sequencing an agent actually gets wrong, learned by doing the work, not farmed from benchmarks or paraphrased from docs. Each one is then refined and scrubbed for general release.
The result is a small, house-standard SKILL.md that stays thin. The wiki-backed ones pull current facts from a live, maintained Agent Wiki over MCP; the rest are self-contained. Tiny context cost, always current. Included with Pro.
What a skill looks like
The discipline, not the docs — the sequencing and gotchas that only surface once you've actually done the work. A live preview of two skills; the full SKILL.md and install are in Pro.
blender-mcppowered by the Blender wikiUse this skill when the user wants Hermes to set up, troubleshoot, or control Blender through the Blender MCP addon, especially from WSL while Blender runs on Windows.
Live reference: this skill is backed by the Blender Agent Wiki — query it over the
agentwikisMCP (search/read_document) for current, detailed facts instead of relying on memory.
Trigger conditions
Load this skill when the user asks to:
- Use Hermes with Blender
- Install or configure the Blender MCP addon
- Create, edit, animate, inspect, screenshot, or render a Blender scene
- Run Blender Python (
bpy) from Hermes - Debug Blender MCP connection issues on port
9876
llm-trainingpowered by the Unsloth wikiMethod-selection discipline for training and fine-tuning large language models. Six interconnected approaches:
| Method | Library | Best For |
|---|---|---|
| PEFT | HuggingFace PEFT | Parameter-efficient fine-tuning with 25+ methods |
| Unsloth | Unsloth | 2-5x faster LoRA/QLoRA with less VRAM |
| TRL | HuggingFace TRL | RLHF, DPO, PPO, GRPO reward modeling |
| GRPO | TRL (GRPO trainer) | Reasoning and task-specific RL fine-tuning |
| Axolotl | Axolotl | YAML-driven multi-method fine-tuning |
| PyTorch FSDP | PyTorch | Distributed training across multiple GPUs |
This skill is deliberately thin. It carries the method-selection discipline and points to live, authoritative sources — it does not vendor a stale copy of anyone's docs. Pull current API details at the point of need from each project's official documentation (linked per section) or via the
agentwikisMCP (search/read_document) where a maintained wiki covers the library.
Section 1: Parameter-Efficient Fine-Tuning (PEFT)
Fine-tune 7B-70B models with <1% of parameters using LoRA, QLoRA, and 25+ other methods. HuggingFace's official library, integrated with the tra
The skills
10 skills authored in real Hermes Agent work, each backed by a maintained wiki (or self-contained). The full SKILL.md + install unlock with Pro.
Control a running Blender instance from Hermes via the Blender MCP addon/socket. Use for Blender scene creation, 3D modeling, materials, lighting, animation, viewport screenshots, renders, and bpy Python automation.
dgx-spark-generative-media →Install, validate, and operate local image/video generation stacks such as ComfyUI on NVIDIA DGX Spark / GB10 (aarch64, unified memory), including model-fit assessment and ARM-specific troubleshooting.
hermes-cron-automation →Build, schedule, and verify Hermes Agent cron automations with pre-run scripts, wakeAgent gates, no-agent watchdogs, and chained handoffs.
hermes-multi-profile-ops →Operate and troubleshoot Hermes multi-profile workflows: Kanban boards, cron jobs, gateway runtime, profile env/toolsets, and paper-only demo loops.
hermes-native-tool-authoring →Author, wire, and verify native Hermes Agent tools in the source tree: tool module, registry schema, toolsets.py exposure, credentials, restart, and validation.
kanban-workflows →Use when operating Hermes Kanban systems as either orchestrator or worker: decomposition, lane discipline, worker lifecycle, and handoff quality.
llm-inference →Run, serve, and evaluate LLMs: local GGUF inference via llama.cpp, high-throughput serving via vLLM, and benchmark evaluation via lm-eval-harness.
llm-training →Train and fine-tune LLMs: LoRA/QLoRA via PEFT and Unsloth, RLHF/DPO/GRPO via TRL and Axolotl, distributed training via PyTorch FSDP. Covers parameter-efficient methods, reward modeling, and distributed scaling.
markdown-kb-publishing →Build local-first Markdown knowledge-base browsers and read-only MCP servers.
unreal-engine-mcp-workflows →Inspect, build, reset, and verify Unreal Engine projects through the Unreal MCP server, including asset-registry discovery, batch scene automation, safe destructive edits, viewport capture, and local Unreal knowledge-base research.
// all skills are included with Pro — $9.99/mo, alongside the XL wikis.