Skip to content

0xdewy/skills

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Skills

AI behavior packs that coding agents can load for specialized roles — spawning a dialectic council to debate an implementation, or a Playwright tester that exercises every feature of an app without writing a single assertion. Skills are closer to workflows than prompt templates: they define phases, roles, and completion signals, not just what to say.


Install

First, install the skill CLI to your system:

./install.sh

This copies skill to a directory on your PATH so you can use it from anywhere.

skill install-all            # install every skill in this repo
skill install <path|url>     # install one skill from a local dir or git URL
skill remove  <name>         # remove one
skill list                   # list installed skills
skill update  <name>         # pull latest for git-cloned skills

See INSTALL.md for Python and system requirements.


Every skill declares when it activates and when it stays silent:

description: Does X. TRIGGER on: "audit my design", "fix the UI".
             SKIP on: "performance issues", "add a backend endpoint".

When your message matches a TRIGGER phrase, the agent loads that skill. When it matches a SKIP phrase, the agent ignores it. That's the idea.

You: "audit my app's design"
Agent: [loads frontend-ux-designer skill]
Agent: "Evaluating layout consistency, typography, color, spacing..."

The Skills

Orchestration — skills that spawn and coordinate multiple sub-agents:

Skill What it does
implementer Implement → blind Skeptic pass + scored Review → Fix loop until 10/10 quality
one-shot-project Full project builder: Architect, PM, parallel implementers, tester
code-smellz Parallel bug-hunter, simplifier, optimizer, and security auditor
brainstormers Dialectical thinker sub-agents debate ideas until convergence
dialectic-council Multi-model debate: Claude + DeepSeek, OpenAI, Gemini, MiniMax

Quality — skills that demand excellence through adversarial refinement:

Skill What it does
student-counsel Student works, philosophers review in dialectic rounds until consensus
look-for-flaws Submit work to a dialectic of sub-agents who find flaws in the codebase

Frontend & Testing — browser-based QA and design auditing:

Skill What it does
frontend-twerkin Starts the app, tests every feature with Playwright, auto-fixes failures
frontend-ux-designer Design audit: layout, typography, color, spacing; produces critique + fixes

Knowledge & Reference — skills that bring deep domain expertise:

Skill What it does
teach-me Learn any codebase with Mermaid diagrams
rust-evm EVM internals: revm, Foundry, bytecode, Yul, gas optimization
prediction_markets Polymarket and Kalshi: APIs, CLOB trading, auth, order types
web-scraping Playwright, httpx, anti-bot evasion, pagination, structured extraction

Meta — skills that create, curate, and constrain other skills:

Skill What it does
skill-creator Create, test, and publish new skills
stakeholder-of-last-resort Speaks for absent parties facing foreseeable harm in decisions

Design Principles

A few things I've found useful when building skills:

Composable. Skills can call skills. Every skill emits a DONE: completion signal so orchestrators know when a sub-agent has finished and what it produced. One skill can spawn another — code-smellz dispatches four parallel auditors, each of which is itself a skill-like role.

Dialectic. Structured disagreement often surfaces what agreement misses. Several skills use adversarial review — sub-agents that genuinely argue with each other — to surface flaws a single pass would miss. student-counsel won't accept work until a council of philosophers reaches consensus that it is beautiful. implementer includes a blind Skeptic pass that reviews the output without seeing the original question — flaws that survive context collapse under it.

Self-critical. Several skills — skill-creator, code-smellz, student-counsel — include mechanisms for revising not just the output but the process itself. A skill that can't improve itself misses its own potential.


Creating a Skill

Skills are SKILL.md files with YAML frontmatter that declare their name, description, and activation contract:

---
name: my-skill
description: What it does. TRIGGER on: "these phrases". SKIP on: "those phrases".
---

## Phase 0: Understand the problem
## Phase 1: Do the work
## Phase 2: Verify

DONE: <output> — <summary>

To create one properly, use the skill that knows how:

Use the skill-creator skill

Or read skills/skill-creator/SKILL.md directly.


MIT License. See LICENSE.

About

A collection of useful skills

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors