Skip to content

rolandbrecht/agent-skills

Repository files navigation

Agent Skills

CI License: MIT

A collection of shareable, high-quality AI coding agent skills and workflows. These skills provide structured guidance, recipes, and helper scripts to make LLM agents significantly more effective at complex codebase tasks.

Why do agents need specialized skills?

While LLM agents have deep semantic understanding, they are fundamentally limited by their context windows and token constraints.

For example, when asked to "rename an API across the project" or "find all dead code":

  • Without skills: An agent must use error-prone text search (grep) which breaks on comments or formatting, or painstakingly open and read hundreds of files sequentially, burning through tokens and forgetting earlier context.
  • With skills (like ast-code-graph): The agent is taught to use industrial-grade tools like ast-grep to instantly perform mathematically sound, structural refactoring across thousands of files in a single step, while using deterministic graph builders to trace global dependencies without reading every file.

These skills provide the industrial tools that let an agent execute systemic changes safely, instantly, and globally.

Skills

AST parsing and code graph indexing for structural code search, refactoring, dead-code detection, dependency tracing, impact analysis, and safe symbol renaming.

Primary tool: ast-grep for structural search and rewrite, with custom scripts for code graph analysis.

Files:

  • SKILL.md — Main skill instructions
  • references/ast-grep-cheatsheet.md — ast-grep CLI and pattern reference
  • references/graph-schema.md — Code graph node/edge schema
  • references/query-patterns.md — Query patterns cookbook
  • scripts/parse-js.mjs — Cross-platform JS/TS AST parser
  • scripts/build-graph.py — Python code graph builder
# Install the ast-code-graph skill
npx skills add rolandbrecht/agent-skills --skill ast-code-graph

Use when encountering any bug, test failure, or unexpected behavior. This skill forces the agent to use the scientific method—tracing root causes and injecting boundary telemetry—before attempting any code fixes. It includes a strict "3 Strikes" escalation protocol to prevent architecture-breaking guesswork.

Files:

  • SKILL.md — The 3-phase systematic debugging framework
  • references/bug-report.md — The 3-strike escalation template
  • references/tracing.md — Actionable boundary-logging patterns (frontend, backend, DB, CI)
# Install the systematic-debugging skill
npx skills add rolandbrecht/agent-skills --skill systematic-debugging

A "meta-skill" executed at the end of complex tasks. It solves the LLM "stateless learner" problem by forcing the agent to codify its hard-won lessons into reusable YAML/Markdown rules before it is allowed to close a task.

Files:

  • SKILL.md — The continuous improvement protocol
  • references/lesson-template.md — Checklists for adding architectural gotchas to project instruction files
# Install the self-reflection skill
npx skills add rolandbrecht/agent-skills --skill self-reflection

Use this skill for any structural code search or replacement task — it is the right choice whenever grep/sed would feel clunky or risky. Ideal for cross-language refactoring, Python 2 → 3 migrations, and modernizing Python 3 idioms.

Files:

  • SKILL.md — Main skill instructions
  • references/hole-syntax.md — Full comby hole type reference
  • references/language-matchers.md — Supported comby language matchers
# Install the structural-refactor skill
npx skills add rolandbrecht/agent-skills --skill structural-refactor

Project Context Files

Optional entry-point files for your project root. They provide a quick decision tree and key rules so the agent knows when to trigger which skill.

File Platform
CLAUDE.md Claude Code
AGENTS.md OpenAI Codex / Generic Agents
llms.txt Web / any LLM (llmstxt.org standard)

Copy the entry-point file for your platform to your project root. They reference skills by name, which agents automatically discover from their skills directory.

Installation

Via skills.sh (Recommended)

The easiest way to install these skills is using skills.sh.

# Install all skills
npx skills add rolandbrecht/agent-skills

# Example: Install only the ast-code-graph skill
npx skills add rolandbrecht/agent-skills --skill ast-code-graph

Manual Installation

You can also install skills manually by copying the skill directories into your platform's skills location.

Claude Code:

# User-level (available in all projects)
cp -r * ~/.claude/skills/

# Or project-level (shared via git)
mkdir -p .claude/skills
cp -r * .claude/skills/

Gemini CLI / Antigravity:

mkdir -p ~/.gemini/antigravity/skills
cp -r * ~/.gemini/antigravity/skills/

OpenAI Codex:

# User-level
mkdir -p ~/.codex/skills
cp -r * ~/.codex/skills/

License

MIT

About

A collection of shareable, high-quality AI coding agent skills and workflows. These skills provide structured guidance, recipes, and helper scripts to make LLM agents significantly more effective at complex codebase tasks.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors