chogos-skills

claude-skills

Claude Code skills for infrastructure, languages, and development best practices.

Installation

# Add the marketplace
claude plugin marketplace add Chogos/claude-skills

# Install the plugin
claude plugin install chogos@chogos-skills

# Update to latest
claude plugin marketplace update chogos-skills
claude plugin update chogos@chogos-skills

Once installed, skills are available as slash commands:

/writing-python
/developing-backend-services

Skills

Skill	Description
developing-backend-services	API design, database patterns, observability, 12-factor
developing-frontend-apps	Component architecture, performance, accessibility
developing-helm-charts	Helm chart conventions, security, values design
managing-k8s-operators	K8s operators: Strimzi, Keycloak, Prometheus, certs
writing-aws-infrastructure	IAM, CloudFormation/CDK, VPC, Lambda, ECS, cost
writing-dockerfiles	Dockerfile conventions, multi-stage builds, security
writing-golang	Go project layout, error handling, concurrency, testing
writing-python	Python project structure, type hints, pytest, async
writing-rego-policies	OPA/Rego policy rules, testing, Kubernetes admission
writing-rust	Rust ownership, error handling, async, traits, testing
writing-shell-scripts	Bash scripting, CLI tools, deployment scripts
writing-postgres	Schema design, queries, migrations — PostgreSQL
writing-sqlite	Schema design, queries, migrations — SQLite

Structure

Each skill uses progressive disclosure — SKILL.md is the overview loaded on trigger, with deeper reference files loaded only when needed:

skills/
└── <skill>/
    ├── SKILL.md              # Overview + conventions (loaded on trigger)
    ├── patterns/             # Full examples by use case (loaded on demand)
    │   ├── <pattern>.md
    │   └── ...
    └── <reference>.md        # Cheatsheets, matrices (loaded on demand)

Naming convention

Skill directories use gerund form: <verb>ing-<noun> (e.g., developing-helm-charts).

Skill authoring best practices

Full reference: Skill authoring best practices — Claude API Docs

Core principles

• Concise is key. The context window is a shared resource. Only add context Claude doesn't already have — challenge every paragraph: "Does Claude really need this?"
• Set appropriate degrees of freedom. High freedom (text instructions) for heuristic tasks, low freedom (exact scripts) for fragile operations. Match specificity to the task's fragility.
• Test with all models you plan to use. What works for Opus might need more detail for Haiku.

Structure constraints

• SKILL.md: Under 500 lines. Overview loaded on trigger — keep it concise with progressive disclosure to pattern files.
• Pattern files (patterns/*.md): 100–250 lines each. Full examples organized by use case.
• Reference files (*.md at skill root): Cheatsheets, matrices, lookup tables.
• One level deep. All reference files link directly from SKILL.md — no nested chains of references.
• TOC for long files. Reference files over 100 lines get a table of contents at the top.

SKILL.md frontmatter

• name: Lowercase letters, numbers, hyphens only. Max 64 chars. Gerund form preferred (writing-python, developing-helm-charts).
• description: Max 1024 chars. Third person. Include both what the skill does and when to use it. Be specific — Claude uses this to pick the right skill from 100+ candidates.

Style

• Code examples over prose. Every guideline should have a snippet.
• Bad/good pairs for anti-patterns. Show what's wrong and the fix.
• No duplicate content between SKILL.md and pattern files. SKILL.md summarizes, patterns go deep.
• Consistent terminology — pick one term and use it throughout (not "endpoint" in one place and "route" in another).
• No time-sensitive information. Use an "old patterns" section with <details> for deprecated approaches.

Workflows and feedback loops

• Break complex operations into clear sequential steps with a checklist Claude can track.
• Use feedback loops: run validator → fix errors → repeat. This pattern greatly improves output quality.
• For conditional workflows, guide Claude through decision points explicitly.

Adding content