2026 Hermes Agent Skills
Advanced Guide

Unlike one-shot system prompts, Hermes treats skills as standardized, evolvable, cross-session procedural memory. For installation see the step-by-step install guide; for persistence layers see the three-layer memory article. This post goes straight into advanced territory.

Every Hermes Skill follows the agentskills.io open standard so the same file works across Hermes, Claude Code, and Cursor. Validate compliance with skills-ref validate ./my-skill.

---
name: my-skill
description: |
  Use when the user needs to [...].
  Handles [...] and [...].
version: 1.0.0
license: MIT
compatibility: Requires git, docker
allowed-tools: Bash(git:*) Read
metadata:
  hermes:
    tags: [devops, automation]
    category: software-development
    related_skills: [github-pr-workflow, test-driven-development]
    requires_toolsets: [terminal]
    fallback_for_toolsets: [web]
---

# My Skill Title

## Overview
## When to Use
## Procedure
## Common Pitfalls
## Verification Checklist

Recommended directory layout:

~/.hermes/skills/
└── my-category/
    └── my-skill/
        ├── SKILL.md              # core steps, target ≤500 lines
        ├── references/           # API docs, loaded on demand
        ├── templates/            # reusable templates
        └── scripts/              # scripts the agent can execute

Skill Bundles are a 2026 Hermes feature: lightweight YAML files that pack multiple skills behind a single slash command. Running /bundle-name loads every listed skill at once. File location: ~/.hermes/skill-bundles/<slug>.yaml.

name: backend-dev
description: |
  Full backend feature workflow — code review, TDD, and PR management.
skills:
  - github-code-review
  - test-driven-development
  - github-pr-workflow
instruction: |
  Always write failing tests first before implementation.
  Never push directly to main.

Bundle priority rules: when a Bundle and a single Skill share a name, the Bundle wins; missing skills are skipped silently; Bundles do not alter the system prompt, so Prompt Cache stays valid. Quick CLI creation:

hermes bundles create backend-dev \
  --skills github-code-review,test-driven-development,github-pr-workflow \
  --instruction "Always write failing tests first"

Conditional activation auto-shows or hides skills based on tools available in the current session. Configure under metadata.hermes:

Classic pattern: after setting FIRECRAWL_KEY or BRAVE_SEARCH_KEY, paid web_search activates and the DuckDuckGo skill disappears via fallback_for_tools: [web_search]. When the API is unavailable, the fallback resurfaces automatically. Platform-aware example: requires_toolsets: [messaging] with platforms: [telegram, discord]—the hermes skills TUI lets you toggle skills per platform independently.

Official install channels:

hermes skills install official/research/arxiv
hermes skills install https://example.com/SKILL.md --name my-skill
hermes skills install github:openai/skills/k8s
hermes skills tap add github:my-org/my-skills
hermes skills tap update
hermes skills tap list

Publishing a team Tap: Create a GitHub repo, control Hub categories with skills.sh.json, and have teammates run hermes skills tap add github:your-org/your-skills-tap; add --token $GH_TOKEN for private repos. Version-control ~/.hermes/skills/ in Git for cross-device sync.

GEPA + DSPy self-evolution (hermes-agent-self-evolution): no weight fine-tuning—analyze execution traces, generate variants, and run multi-objective Pareto optimization on skill text. Each optimization run costs roughly $2–10 in API fees with no GPU. Five stages: ① execution trace collection (SQLite); ② reflective failure analysis; ③ targeted mutation (10–20 SKILL.md variants); ④ multi-objective Pareto evaluation (success rate × token efficiency × speed); ⑤ human PR review.

export HERMES_AGENT_PATH=~/.hermes
python -m evolution.skills.evolve_skill \
    --skill github-code-review \
    --iterations 10 \
    --eval-source sessiondb

python -m evolution.skills.evolve_skill \
    --skill github-code-review \
    --iterations 10 \
    --eval-source mixed \
    --trace-dirs ~/.claude/traces,~/.hermes/sessions

Four safety guardrails: full pytest suite must pass 100%; Skills ≤ 15KB; Prompt Cache compatibility; semantic preservation checks. Official evolution roadmap: Phase 1 Skill files (done) → Phase 2 tool descriptions → Phase 3 system prompt → Phase 4 tool implementation code → Phase 5 fully automated loop.

Plugin skills load under the plugin:skill namespace—they do not appear in default skills_list and activate only on explicit user call; plugin skills can reference each other. Advanced authoring: write trigger conditions in description; put concrete failure modes and fixes in Common Pitfalls; split large skills into references/ (>500 lines recommended split, >15KB exceeds GEPA limit); the skill_manage tool lets the agent maintain skills dynamically—set agent_writes_require_approval: true in config.yaml for a human approval gate.

Build a blog-workflow Bundle packing seo-keyword-research, outline-generator, code-example-validator, bilingual-checker, and publish-to-platform. The bundle instruction requires SEO keyword research first, runnable code examples, and bilingual titles. A custom seo-keyword-research skill sets requires_toolsets: [web] in metadata.hermes; its Procedure covers English long-tail patterns (how to X, X vs Y) and cross-validates against platform trend data.

Closing a laptop lid, running on an x86 VPS without native macOS paths, or keeping skills on unversioned local disk all interrupt skill compounding and GEPA trace accumulation. For production workloads that need a 24/7 Gateway, skill self-evolution, and cross-device sync, MESHLAUNCH Mac Mini M4 bare-metal cloud rental is the practical fit: dedicated Apple Silicon, native launchd daemons, flexible daily/weekly/monthly billing, and the closest match to Hermes official macOS paths. See pricing, help center, or order a cloud Mac. Further reading: Hermes docs, GEPA algorithm, DSPy framework.