synit.io - Agent Skill Collection

A collection of agent skills created and maintained by synit.io. Each skill packages instructions, scripts, and reference material that let an AI agent operate a specific tool or service safely and predictably.

About synit.io

synit.io is a German consultancy for data engineering, analytics, and AI, working with mid-market companies in the DACH region under the principle of "Owned Intelligence": the AI is the tool, the data is the plan. Focus areas include data infrastructure, machine learning for forecasting and anomaly detection, language-model automation, and system integration.

The skills in this repository grew out of that work. They are published so that other teams can reuse them with their own agents and tenants.

What is in this repository

Each top-level directory is one skill. A skill follows the open Agent Skills format:

<skill-name>/
  SKILL.md        # frontmatter (name, description) plus operating instructions
  scripts/        # CLI tooling the agent runs
  references/     # API notes, config templates, background material
  tests/          # regression tests for the scripts
  agents/         # optional harness metadata, e.g. openai.yaml for Codex

Skills cover different use cases, tools, and services. They are independent of each other; install only the ones you need.

Skills

Skill	Description
docker-agent-builder	Create, edit, migrate, repair, review, and validate Docker Agent YAML. Includes current-schema validation, semantic and secret checks, source-cache corruption checks, runtime dry-run support, and ready-to-adapt templates.
m42sd-skill	Operate Matrix42 Enterprise Service Management through the m42Services API: ticket, journal, user, knowledge-base, and service-catalog workflows for helpdesk agents. Includes a stateless Python CLI, guided tenant setup, and safety rules for mutations such as closing or forwarding tickets.
nexthink-campaign-translator	Add languages to a Nexthink campaign export JSON. The agent translates question and choice text; a Python helper writes the translation overlays, verifies that question IDs, {{placeholders}}, and the HTML structure (tags, attributes, href values) match the source, updates multiLanguageInfo, and repairs INVALID overlays. Supported languages come from a configurable allowlist.

Installation

Every skills-compatible agent scans one or more directories for <skill-name>/SKILL.md. Installing a skill means placing (or symlinking) the skill folder into such a directory. You can let the skills CLI do that for you (next section) or do it by hand (the sections after). The manual steps assume you cloned this repository first:

git clone https://github.com/synit-io/skills.git
cd skills

Symlinking keeps the installed skill in sync with git pull; copying gives you an isolated snapshot. Both work. Skill scripts keep credentials and tenant config outside the skill folder (for example ~/.config/m42sd/m42_config.json, or the path in M42_CONFIG_PATH), so reinstalling or updating a skill does not touch them.

After installing, open the skill's SKILL.md and follow its setup section (credentials, tenant discovery, safety rules) before the first real task.

Quick install with the skills CLI

The fastest route on any supported harness is the open-source skills CLI behind skills.sh. It detects the agents installed on your machine and copies the skill into the right directory for each.

Install every skill in this repository into the current project:

npx skills add synit-io/skills

Install one skill, for your user account, for specific agents:

npx skills add synit-io/skills --skill m42sd-skill -g -a claude-code -a codex -a opencode

List what the CLI would install without installing:

npx skills add synit-io/skills --list

The CLI copies files rather than symlinking. Re-run the same command to pull a newer version. A m42_config.json created by an older m42sd-skill version lives inside the installed copy (scripts/m42_config.json) and is still read from there; move it to ~/.config/m42sd/ before reinstalling so it is not overwritten. The CLI sends anonymous install telemetry to skills.sh; set DISABLE_TELEMETRY=1 to opt out.

Where each harness looks

Harness	Project-level	User-level	Invoke
Claude Code	.claude/skills/	~/.claude/skills/	/skill-name or automatic
Codex (CLI, IDE)	.agents/skills/	~/.agents/skills/	$skill-name, list with /skills
OpenCode	.opencode/skills/ (also reads .claude/skills/, .agents/skills/)	~/.config/opencode/skills/ (also reads ~/.claude/skills/, ~/.agents/skills/)	skill tool, automatic
Gemini CLI	.gemini/skills/ or .agents/skills/	~/.gemini/skills/ or ~/.agents/skills/	/skills commands, automatic
Cursor	.cursor/skills/ or .agents/skills/ (also reads .claude/skills/)	~/.cursor/skills/ or ~/.agents/skills/ (also reads ~/.claude/skills/)	/ in Agent chat, automatic

.agents/skills/ is the cross-tool convention. One install into ~/.agents/skills/ is picked up by Codex, OpenCode, Gemini CLI, and Cursor. Claude Code uses ~/.claude/skills/, which OpenCode and Cursor also read.

Claude Code

Personal install, available in every project:

mkdir -p ~/.claude/skills
ln -s "$(pwd)/m42sd-skill" ~/.claude/skills/m42sd-skill

Project install, checked into a repository for the whole team:

mkdir -p .claude/skills
cp -r /path/to/skills/m42sd-skill .claude/skills/

Claude Code picks up changes to SKILL.md within the running session. Invoke with /m42sd-skill, or let Claude select the skill from its description.

Codex

Codex reads .agents/skills/ in the working directory, its parents up to the repository root, and ~/.agents/skills/ for personal skills:

mkdir -p ~/.agents/skills
ln -s "$(pwd)/m42sd-skill" ~/.agents/skills/m42sd-skill

Mention a skill with $m42sd-skill, or run /skills to see what is loaded. Restart Codex if a new skill does not show up. To disable a skill without deleting it, add a [[skills.config]] entry in ~/.codex/config.toml.

OpenCode

OpenCode reads its own directories plus the Claude Code and .agents paths, so an existing Claude Code or Codex install already works. For a dedicated install:

mkdir -p ~/.config/opencode/skills
ln -s "$(pwd)/m42sd-skill" ~/.config/opencode/skills/m42sd-skill

Project-local skills go in .opencode/skills/. Skills load through the native skill tool. Control access in opencode.json:

{
  "permission": {
    "skill": {
      "*": "allow"
    }
  }
}

Other agent harnesses

Any tool that implements the Agent Skills format (Gemini CLI, Cursor, GitHub Copilot, Goose, OpenHands, and others listed at agentskills.io) follows the same pattern:

1. Find the skill directory your harness scans. Check its documentation; most accept .agents/skills/ in the project and ~/.agents/skills/ in your home directory.
2. Copy or symlink the skill folder there so that <skills-dir>/<skill-name>/SKILL.md exists.
3. Restart or reload the harness and confirm the skill is listed.
4. Make sure the harness can run shell commands and that python3 is on the path. Check the selected skill's SKILL.md for any additional Python packages (only docker-agent-builder needs some).

If your harness has no skill support, paste the contents of SKILL.md into the agent's system prompt or its AGENTS.md, keep the scripts/ and references/ folders reachable from the working directory, and give the agent shell access. The instructions reference scripts as <skill-dir>/scripts/...; tell the agent which directory <skill-dir> stands for.

Updating

cd /path/to/skills
git pull

Symlinked installs update immediately. Copied installs need the copy step again. Installs made with the skills CLI update by re-running the same npx skills add command.

Security

Skill scripts that need credentials store them in local config files with owner-only permissions, outside the skill folder and outside version control. Pass secrets through environment variables or interactive prompts, never as command-line arguments. Never commit tokens, tenant profiles, or discovery output, and never install a skill into a directory that is committed to a public repository together with its generated config. The same applies to customer data such as campaign exports and translation bundles: keep them out of the repository.

Contributing

Issues and pull requests are welcome. Keep each skill self-contained, add or update its tests, and do not include tenant-specific values or secrets.

License

All skills in this repository are released under the MIT License. Bundled upstream Docker material in docker-agent-builder/references/ retains its original Apache License 2.0 text and notices (THIRD_PARTY_NOTICES.md, docker-agent-LICENSE.txt).