49 custom-built skills | 6 categories | 2 repos | Quality gate for every skill

All skills are custom-built — developed and maintained by AI Engineering. They run as Claude Code auto-load skills in .claude/skills/ directories.

What Are Skills?

Skills are reusable prompt templates for Claude Code. Each skill consists of a SKILL.md file with YAML frontmatter (metadata) and a body (instructions). Claude Code loads them automatically when you work in the corresponding directory.

Official Anthropic standard + custom extensions:

  • Anthropic fields: name, description, instructions, command_injection
  • Our extensions: version, requires, produces, error-refs, learning-refs
  • String substitutions: $ARGUMENTS, ${CLAUDE_SKILL_DIR}, ${workspaceFolder}

Skill Repos

Repo Focus Skills Categories
interne Repo Infrastructure, agents, voice gateway 25 Infra, agent, content, meta
Playbook01 Business, marketing, organization 24 Marketing, firma, operations

Category Overview

Infrastructure & Ops (6)

Skill Trigger Description
deploy deploy, VG, dashboard Deploy VG/Dashboard to Docker Swarm
pve-operations VM, Proxmox, migrate Proxmox VM management
swarm-recovery services down, recover Docker Swarm service recovery
swarm-raft-recovery raft, quorum loss Nuclear: Raft quorum loss recovery
n8n-workflow-ops n8n, workflow, RSS n8n workflow CRUD and management
version-bump version, bump, release Atomic version bump

Agent Management (7)

Skill Trigger Description
jim-manager delegate, team status Team delegation and tracking
agent-vault credentials, tokens Central credential database
agent-task task create ERPNext task management
mm-wait Manager-Agent polling Message Bus polling for Manager-Agent
mm-wait-echo Logger-Agent polling Message Bus polling for Logger-Agent
mm-wait-jim01 Developer-Agent polling Message Bus polling for Developer-Agent
echo-log-learning teach echo_log Learning sessions for echo_log

Documentation & Orchestration (4)

Skill Trigger Description
doc-updater update docs, sync Update docs after changes
docs-navigator find doc, where is Navigate across 97+ docs
echo-log-context stack overview Instant stack context
full-sync full-sync, update all 10-step sync pipeline

Content & Creative (5)

Skill Trigger Description
kroki-diagrams diagram, mermaid Diagrams via Kroki
content-quality content review Content quality gate
comfyui-workflows comfyui, flux, image ComfyUI workflow design
dashboard-verifier dashboard verify Ops dashboard health check
sigma-upgrade sigma, network graph Sigma.js force-atlas upgrade

PDF & Output (3)

Skill Trigger Description
pdf-enterprise-lite PDF, report, lightweight PDF lite for free products
pdf-enterprise-pro PDF, branded, pro PDF pro for paid products
pdf-dark-theme-generator pdf, dark theme Dark-theme PDFs via WeasyPrint

Meta (2)

Skill Trigger Description
skill-router which skill, routing Route requests to right skill
skill-creator new skill Scaffold new skills

Marketing (10)

Skill Trigger Description
content-calendar blog, content plan Blog content pipeline
social-media-pipeline social media, post Multi-platform posting
brand-consistency brand check, before publish Brand guidelines check
market-copy copywriting, headlines Sales copy generation
market-seo seo, keywords, meta SEO analysis + optimization
market-ads ads, werbung, campaign Ad copy + strategy
market-funnel funnel, conversion Sales funnel analysis
market-landing landing page, CRO Landing page CRO
market-proposal proposal, quote Customer proposals
market-audit marketing audit Marketing audit (score 0-100)

Organization & Firma (5)

Skill Trigger Description
firma-okr OKR, goals OKR tracking
firma-meeting meeting, standup Meeting notes
firma-sprint sprint, planning Sprint planning
firma-finance cashflow, revenue Financial overview
firma-onboarding onboarding, new agent Agent onboarding

Operations (7)

Skill Trigger Description
email-assistant email, triage Email triage + drafts
dr-recovery disaster recovery DR quick reference
weekly-review weekly review Weekly review
competitor-watch competitor, market Competitor scan
multi-shop-manager shop, listing Multi-platform shop
notebooklm-audio notebooklm, audio Audio guide production
task-manager (deprecated) Replaced by ERPNext tasks

Quality Gate

Every skill must meet these requirements:

  • SKILL.md with complete YAML frontmatter
  • name and description set
  • At least 1 trigger keyword defined
  • Body with clear instructions (no filler)
  • Tested in the respective repo context
  • Documented in skill catalog

Skill Lifecycle

CREATE → Idea + /skill-creator → scaffold SKILL.md
  ↓
DEVELOP → Write body, set frontmatter
  ↓
TEST → Execute in context, check result
  ↓
CATALOG → Register in skill catalog
  ↓
MAINTAIN → Bump version, maintain error/learning refs
  ↓
ARCHIVE → Set deprecated flag, document replacement

External Skills

Currently no external skills installed. Evaluated sources:

Source Skills Status
alirezarezvani/claude-skills 169 skills Evaluated
anthropics/skills 6 official Evaluated
openPawDev/pawmode 38 skills Research

Skill Architecture in Detail

Frontmatter Fields (Complete Reference)

Field Type Required Description
name String Yes Unique ID (kebab-case)
description String Yes Brief description with trigger keywords
version String Yes Semantic versioning (X.Y.Z)
requires Array Yes Dependencies (rules, skills, tools)
produces Array Yes Output types (file, notification, etc.)
disable-model-invocation Boolean No true = manual invocation only
user-invocable Boolean No false = Claude invokes only
model String No haiku|sonnet|opus
context String No fork = isolated subagent context
agent String No Explore|Plan|general-purpose
allowed-tools Array No Tool whitelist
error-refs Array No References to ERROR_REGISTRY.md
learning-refs Array No References to LEARNINGS_REGISTRY.md
last-verified String Yes Last verification date (YYYY-MM-DD)

String Substitutions (Dynamic Values)

Claude Code replaces these at runtime:

Variable Description Example
$ARGUMENTS All user input after skill name /skill-name arg1 arg2arg1 arg2
$0 First argument arg1
$1 Second argument arg2
$N N-th argument $3 for third argument
${CLAUDE_SESSION_ID} Unique session identifier abc-def-ghi-jkl
${CLAUDE_SKILL_DIR} Absolute path to skill directory /Users/name/.claude/skills/skill-name/

Dynamic Context Injection

Shell commands run before skill executes with !`command`! syntax:

---
command_injection: "git log --oneline -5"
---

## Current Status
!`curl -s http://api.example.com/status | jq .`!

This value gets inserted before Claude sees the body.

Useful for:

  • Current system states
  • Git history
  • Deployment status
  • API data

Bundled Skills (Standard)

These skills come with Claude Code:

Skill Description Usage
/batch Parallelize large changes across worktrees /batch migrate src/ from React to Vue
/claude-api Load Claude API reference (Python, TypeScript, etc.) Auto-load on import anthropic
/debug Debug current session (logs, errors) /debug why is this failing?
/loop Repeat prompt on interval /loop 5m check if deploy finished
/simplify Review code for reuse and efficiency /simplify focus on memory usage

Skill Lifecycle (Extended)

IDEATION
  ↓ (User has skill idea)
CREATE
  ↓ (Create SKILL.md with frontmatter)
DRAFT
  ↓ (Write body, not yet productive)
DEVELOP
  ↓ (Add tests, scripts, templates)
TEST
  ↓ (Manual test in context)
QUALITY_GATE
  ↓ (Pass checklist)
CATALOG
  ↓ (Register in catalog)
PUBLISH
  ↓ (Skill is productive)
MAINTAIN
  ↓ (Monthly — update last-verified)
DEPRECATE
  ↓ (Replace with better alternative)
ARCHIVE
  └ (Move to _archive/)

Skill Best Practices

One skill, one job

Create focused skills:

  • /deploy — Deploy VG
  • /test-suite — Run tests
  • /do-everything — Does everything (too much)

Explicit Trigger Keywords

Description should have clear keywords:

description: "Deploy Voice Gateway or Dashboard — Trigger: deploy, VG, dashboard"

Claude uses keywords for auto-loading.

Document Error Handling

Write top 5 errors and fixes:

## Error Handling

- SSH timeout → Retry after 10s, max 3x
- Build failure → Show logs, don't rebuild
- Health check failed → Manual check needed

No Secrets in Skills

# CORRECT
Load credentials via: vault.py get <service> <key>

# WRONG
token="sk-abc123def456"  ❌ NEVER

Skill Discovery & Auto-Loading

Claude Code:

  1. Loads all skill descriptions into context (so Claude knows what's available)
  2. Loads only the body of a skill when:
    • User executes /skill-name, OR
    • Claude auto-loads based on description

Example:

User: "How does the auth system work?"
Claude sees: description "explain-code: explains code with diagrams"
→ Auto-loads the skill
→ Gives diagram + explanation

Distribute Skills in Plugins

Skills can be distributed as part of plugins:

my-plugin/
├── .claude-plugin/plugin.json
└── skills/
    └── code-review/
        └── SKILL.md

Users install:

/plugin install my-plugin
/code-review-plugin:code-review  # With namespace

Further Reading


Last updated: 2026-03-21 | 49 custom-built skills + Anthropic bundled skills