Skip to content

eea/eea.agent.skills

BranchesTags

Repository files navigation

EEA AI Harness

The canonical organization-wide AI harness for all EEA coding agents.

This repository is the single source of truth for how AI coding agents (OpenCode, Claude Code, Gemini, Copilot, etc.) behave across all European Environment Agency (EEA) projects.

It contains org-wide rules, reusable skills, mandatory protocols, shared knowledge, and tool-specific wiring instructions.

The Initial Owner of the Original Code is European Environment Agency (EEA).

Quick Start

For Developers

Add the EEA harness to your project in 30 seconds. Pick one of the three methods:

A. Automated Global Install (recommended) — one command for all your agents:

# With agentget
agentget install eea/eea.agent.skills

# Or with curl
curl -fsSL https://raw.githubusercontent.com/eea/eea.agent.skills/main/scripts/install.sh | bash

B. Manual Global Install — clone once, symlink per agent:

git clone https://github.com/eea/eea.agent.skills.git ~/.eea/agent-harness
~/.eea/agent-harness/scripts/install.sh --local

C. Project-Embedded — no global setup, reference inside one project:

# OpenCode: remote URL in your project's opencode.json
cat > opencode.json << 'EOF'
{
  "instructions": [
    "https://raw.githubusercontent.com/eea/eea.agent.skills/main/harness/EEA-HARNESS.md"
  ]
}
EOF

Full bootstrap guide: docs/BOOTSTRAP.md

After installing, run the health check to verify everything is wired correctly:

./scripts/verify.sh

For Project Maintainers

Create project-local instructions that add to the org harness:

mkdir -p .agents
cp ~/.eea/agent-harness/templates/dot-agents/AGENTS.md .agents/AGENTS.md
# Edit .agents/AGENTS.md with project-specific rules
cp ~/.eea/agent-harness/templates/dot-agents/opencode.json opencode.json

Repository Structure

ea.agent.skills/
├── harness/
│   └── EEA-HARNESS.md           # ORG-WIDE: canonical harness for all EEA projects
├── AGENTS.md                    # REPO-LOCAL: how to work on THIS repo
├── skills/                      # Distributable merged skills (do not edit directly)
├── src/skills/                  # Source: upstream SKILL.md + EEA-OVERRIDES.md
├── rules/                       # Org-wide prohibitions & mandatory behaviors
├── shared/                      # Cross-project knowledge base
├── instructions/                # Generic org-wide instruction templates
├── workflows/                   # Multi-skill orchestration recipes
├── plugins/                     # Tool-specific adapters (agentget manifest)
├── scripts/                     # Build + install automation
│   ├── build.sh
│   ├── install.sh
│   ├── verify.sh
│   └── validate-skills.sh
├── docs/                        # Documentation and per-tool agent profiles
├── templates/                   # Templates for project-local .agents/ setup
├── catalog.yaml                 # Machine-readable skill index
├── CHANGELOG.md
├── CONTRIBUTING.md
└── LICENSE

Key Design Decision: Two-Layer Harness

Layer File Purpose
Org-wide harness/EEA-HARNESS.md Loaded by ALL EEA projects. Contains routing, prohibitions, mandatory actions, skill references.
Repo-local AGENTS.md (root) Only for this repo. Describes how to maintain the harness itself.
Project-local {repo}/AGENTS.md Per-project rules that add to org harness.

This separation ensures:

  • Org rules are version-controlled and distributed automatically
  • Repo rules don't leak into other projects
  • Project rules can add context without duplicating org standards

The Harness

What It Contains

The canonical org harness (harness/EEA-HARNESS.md) includes:

  1. Context Routing Rules — when to load which skills
  2. Global Prohibitions — security, operational safety, code quality rules that apply everywhere
  3. Mandatory Actions — required behaviors (show git status, propose commit message, ask for confirmation)
  4. Knowledge Accumulation Protocol — how to capture decisions and gotchas
  5. Skill Library Reference — catalog of available skills with usage instructions
  6. Tool-Specific Wiring — how to connect OpenCode, Claude, Gemini, etc.

Distribution Options

Method Best For Command Agents Supported
A. Automated Global Install Most users — one command, all agents agentget install eea/eea.agent.skills or curl | bash OpenCode, Claude, Hermes, Pi, Gemini
B. Manual Global Install Air-gapped or control-oriented users git clone ... && ./scripts/install.sh --local OpenCode, Claude, Hermes, Pi, Gemini
C. Project-Embedded No global setup, harness lives in one project Remote URL, git submodule, or inline copy in project Any agent

Skills

Available Skills

Skill Description Category Upstream Source
docker-expert Advanced Docker containerization, multi-stage builds, security hardening devops sickn33/antigravity-awesome-skills
react-best-practices React/Next.js performance optimization (70+ rules) frontend vercel-labs/agent-skills
composition-patterns React composition patterns, compound components frontend vercel-labs/agent-skills
web-design-guidelines UI review, accessibility, UX audit design vercel-labs/agent-skills
react-native-skills React Native/Expo best practices mobile vercel-labs/agent-skills
react-view-transitions View transitions, animations, shared elements frontend vercel-labs/agent-skills
eea-design-system EEA design system for Volto-based web applications design Self-contained (EEA-specific)

Installing Skills

Option 1: Copy from cloned repo

git clone https://github.com/eea/eea.agent.skills.git
cp eea.agent.skills/skills/docker-expert/SKILL.md ~/.config/opencode/skills/docker-expert/SKILL.md

Option 2: agentget

agentget install eea/eea.agent.skills

Note on GitHub Releases: GitHub Releases distribution is discontinued as of 2026-05-16. Skills are now always installed from source via the install script or agentget. Previous releases and tags have been removed. See CHANGELOG.md for details.

Using Skills

Invoke naturally in any agent:

Use docker-expert to containerize this Python app
Use react-best-practices to optimize this component
Use web-design-guidelines to audit accessibility

Sync Strategy: Two-File Overlay

Each skill follows a two-file overlay pattern:

  1. src/skills/<name>/SKILL.md — Upstream base content (auto-updated from upstream source)
  2. src/skills/<name>/EEA-OVERRIDES.md — EEA-specific customizations
┌─────────────────────────────────────────┐
│              Agent loads                 │
│         SKILL.md + EEA-OVERRIDES.md     │
└─────────────────────────────────────────┘
         ↓                    ↓
┌─────────────────┐  ┌─────────────────────┐
│   Upstream      │  │   EEA Overrides    │
│   (src/skills/) │  │   (src/skills/)    │
└─────────────────┘  └─────────────────────┘
         ↓
┌─────────────────────────────────────────┐
│   Merged output → skills/<name>/        │
│   (agentget discovers from here)        │
└─────────────────────────────────────────┘

Benefits:

  • Upstream changes apply automatically (no merge conflicts in override sections)
  • EEA customizations never lost during upstream updates
  • Clear separation of "what's upstream" vs "what's EEA"

📄 Full sync strategy: docs/SYNC-STRATEGY.md

Adding Skills to the Harness

Adding a New Skill

  1. Create skill directory under src/skills/{skill-name}/
  2. Add upstream SKILL.md (if based on upstream source)
  3. Create EEA-OVERRIDES.md with EEA-specific customizations
  4. Add metadata.json with skill metadata
  5. Update catalog.yaml with new skill entry
  6. Build merged skill: ./scripts/build.sh {skill-name}
  7. Verify: check that skills/{skill-name}/SKILL.md was generated correctly and git status shows the expected changes
  8. Commit: skill: add {skill-name}

Updating an Existing Skill

  1. Sync upstream changes to src/skills/{name}/SKILL.md
  2. Update EEA-OVERRIDES.md if upstream changes affect EEA customizations
  3. Rebuild: ./scripts/build.sh {name}
  4. Update catalog.yaml version if applicable
  5. Commit: skill: update {name} to v{X.Y.Z}

Contributing

See CONTRIBUTING.md for:

  • Adding new skills
  • Updating existing skills
  • Updating the org harness (harness/EEA-HARNESS.md)
  • Adding new agent profiles
  • Sync workflow documentation

Maintaining the Harness

When an agent makes a mistake the harness should have prevented, add a rule instead of rewriting the prompt.

See docs/harness-maintenance.md for the full maintenance philosophy and process.

Commit Convention

Use Conventional Commits:

Type Use For
skill: Adding or updating a skill
harness: Changes to harness/EEA-HARNESS.md or org-wide rules
docs: Documentation updates
build: Build script or CI changes
chore: Maintenance, dependencies, formatting

CI/CD

Validation Workflow

.github/workflows/validate-skills.yml runs on every push:

  • Lint SKILL.md files for structural conformance
  • Check token count (warn if > 500 lines / 5k tokens)
  • Validate catalog.yaml schema
  • Verify skills/ is up-to-date with src/skills/
  • Run agentskills validate (via skills-ref) against all source and merged skills for Agent Skills spec compliance

Harness Validation

.github/workflows/validate-harness.yml runs when harness files change:

  • Verify harness/EEA-HARNESS.md exists and is valid
  • Check all referenced files exist
  • Scan for accidental secrets
  • Validate agent profiles

CHANGELOG Check

.github/workflows/check-changelog.yml runs on pull requests to ensure:

  • Significant changes are documented in CHANGELOG.md
  • Entries follow the established format

License

MIT — See LICENSE

Last updated: 2026-05-21 after skills-ref validation integration and eea-design-system skill addition

About

Centralized repository for agentic skills used across EEA's unified dev workflow with AI coding assistants.

Resources

Readme

License

MIT license

Contributing

Contributing
Activity
Custom properties

Stars

2 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages