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.
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.
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 instructionsreferences/ast-grep-cheatsheet.md— ast-grep CLI and pattern referencereferences/graph-schema.md— Code graph node/edge schemareferences/query-patterns.md— Query patterns cookbookscripts/parse-js.mjs— Cross-platform JS/TS AST parserscripts/build-graph.py— Python code graph builder
# Install the ast-code-graph skill
npx skills add rolandbrecht/agent-skills --skill ast-code-graphUse 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 frameworkreferences/bug-report.md— The 3-strike escalation templatereferences/tracing.md— Actionable boundary-logging patterns (frontend, backend, DB, CI)
# Install the systematic-debugging skill
npx skills add rolandbrecht/agent-skills --skill systematic-debuggingA "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 protocolreferences/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-reflectionUse 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 instructionsreferences/hole-syntax.md— Full comby hole type referencereferences/language-matchers.md— Supported comby language matchers
# Install the structural-refactor skill
npx skills add rolandbrecht/agent-skills --skill structural-refactorOptional 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.
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-graphYou 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/MIT