Skip to content

terion-name/skills

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Skills

My collection of crafted skills.

Install

Skills are published at https://skills.sh

npx skills add terion-name/skills

Or install a single skill:

npx skills add terion-name/skills/writing-good-code

Skills

Code quality

A skillset for guiding AI coding agents to write and maintain code that is simple, readable, and correct — biased against over-engineering, pattern cargo-culting, and the wrong abstractions that calcify over time. Based on professional literature and articles.

Philosophy

Modern, practical, non-religious. Avoids the dogmatic mid-2000s Clean Code style (3-line methods, polymorphism-over-conditionals-always, IFoo-per-Foo) that reads well in books and produces unmaintainable codebases.

Core ideas:

  • Complexity = obscurity + dependencies (Ousterhout).
  • Modules hide secrets, not steps (Parnas).
  • Simple is not-complected; easy is near-at-hand (Hickey).
  • Duplication is far cheaper than the wrong abstraction (Metz).
  • Cognitive load is the real budget; agents must use mechanical proxies for the human "feel" they lack (Zakirullin).
  • Beauty is a correctness heuristic (DHH, Alexander).

Influences: Ousterhout, Beck (Tidy First?), Parnas, Hickey, Fowler, Metz, North (CUPID), Carmack/Muratori on over-abstraction, Alexis King ("Parse, don't validate"), Carson Gross (Locality of Behavior), Gall, Conway, Hyrum, Chesterton, Spolsky (leaky abstractions), Thomson (Postel critique), the Go Proverbs, Rob Pike, DHH, the Grug Brained Developer.

For further details see docs page

Skill Description
writing-good-code Greenfield authoring — naming, module shape, depth, errors, language idioms
refactoring-and-reviewing-code Improving existing code without changing behavior — smells, tidyings, safe sequences
abstraction-quality Decision procedure for when to extract an abstraction and when to duplicate instead

The three code-quality skills interlink: writing-good-code and refactoring-and-reviewing-code both reference abstraction-quality for the extract-vs-duplicate question.

Security

A defender-first skillset for repository security review: threat model first, tool-assisted scan second, validation before severity, and durable artifacts under .security/. It is modeled after the Codex/Aardvark style of scanning: combine deterministic tools with semantic review, then prove or discard findings instead of shipping scanner noise.

Philosophy

Security review should be adversarial without becoming theatrical. The goal is not to produce the longest list of possible weaknesses; the goal is to identify realistic attack paths through the system the user actually runs.

Core ideas:

  • Threat model before checklist; assets, trust boundaries, and reachable entry points decide what matters.
  • Scanner output is evidence, not a finding; every serious issue needs code-path validation.
  • Severity follows exploitability; reachability and impact beat generic CVSS labels.
  • Proof stays local and safe; no external attacks, no secret disclosure, no destructive testing.
  • Reports are working artifacts; .security/ should help the next scan and the remediation pass.

For further details see docs page

Skill Description
security-audit Threat-model-driven security audit, validation, exploit chaining, and .security/ reporting

API architecture

Skill Description
orpc-contract-first oRPC contract-first API design for TypeScript monorepos with contracts in a shared package

Coder Mux agents

Agent definitions for Coder Mux live under agents/ in this repo. They are Markdown persona files with YAML frontmatter. Unlike AGENTS.md, this is a Mux-specific format, not a broadly standardized repository instruction format.

For further details see docs page

Agent Description
Orchestrator Coordinates delegated implementation, integrates child patches, and keeps the parent workspace focused. It was in mux untill it was removed from basic set, alternative provided by @ibetitsmike.
Security Officer Runs threat-model-driven security reviews and writes .security/ scan artifacts

Adding skills manually

The SKILL.md files are plain Markdown and work in any harness. Two patterns:

A) Native skills format — the harness loads the folder directly.

B) Paste into instructions — strip the YAML frontmatter, paste the body into the harness's instruction file. Works everywhere.


Claude Code

Drop the skill folder into your project or user skills directory:

# project-level
cp -r skills/writing-good-code .claude/skills/

# user-level (all projects)
cp -r skills/writing-good-code ~/.claude/skills/

Claude.ai

Upload each skill folder via the Skills UI, or use npx skills add terion-name/skills.

OpenAI Codex CLI

Add to AGENTS.md in the repo root (project-level) or ~/.codex/instructions.md (user-level).

For Coder Mux, copy or symlink the files in agents/ into .mux/agents/ for project-level agents or ~/.mux/agents/ for global agents. The Security Officer agent composes with the security-audit skill: the agent supplies the runnable persona and guardrails, while the skill supplies the methodology, policy references, tooling guidance, and report formats.

Gemini CLI

Add skill bodies to GEMINI.md in the repo root, or to ~/.gemini/GEMINI.md for user-level.

Cursor

Create a rule file for each skill under .cursor/rules/:

# strip frontmatter and write to a .mdc rule file
awk 'BEGIN{f=0} /^---/{f++; next} f>=2{print}' skills/writing-good-code/SKILL.md \
  > .cursor/rules/writing-good-code.mdc

Or paste the body manually into .cursorrules.

Windsurf

Append the skill body to .windsurfrules at the project root:

awk 'BEGIN{f=0} /^---/{f++; next} f>=2{print}' skills/writing-good-code/SKILL.md \
  >> .windsurfrules

Cline / Roo

Paste the skill body into .clinerules at the project root, or add it via the Cline system prompt UI.

GitHub Copilot

Append skill bodies to .github/copilot-instructions.md:

awk 'BEGIN{f=0} /^---/{f++; next} f>=2{print}' skills/writing-good-code/SKILL.md \
  >> .github/copilot-instructions.md

Aider

Pass skill files with --read at startup:

aider --read skills/writing-good-code/SKILL.md

Or add to .aider.conf.yml:

read:
  - skills/writing-good-code/SKILL.md

Continue (VS Code / JetBrains)

Add skill files as context via .continue/config.json using the file context provider, or paste the body into a custom system prompt in the Continue settings.

Structure

skills/
├── <skill-name>/
│   ├── SKILL.md          # agent instructions + trigger description (YAML frontmatter)
│   ├── references/       # supporting docs the agent can pull on demand
│   ├── assets/           # optional templates and worked examples
│   └── scripts/          # optional helper scripts
agents/
├── <agent-name>.md       # Coder Mux agent definition source
└── ...

Each SKILL.md has a YAML frontmatter block with name and description fields. The description controls when the agent decides the skill is relevant.

Agent files also use YAML frontmatter. They are not required to use the skills, and the format is currently specific to Coder Mux rather than the general AGENTS.md convention.

License

MIT

About

Crafted skills mostly for development

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages