Remy Skill Recipes

Engineering-grade skill recipes for working with LLMs.

This repository is a curated collection of reusable, structured prompt workflows designed for software engineering tasks — debugging, documentation, review, research, automation, and system analysis.

Each skill follows the SKILL.md open standard — compatible with Cursor, Claude Code, VS Code + Copilot, Codex, Gemini CLI, and other SKILL.md-compatible agents.

Quick Start

Browse

| Skill | Type | Description | |---|---|---| | architecture-spec | Execution | Risk-based architecture doc generation with auto-leveling (A/B/C) | | branch-context-check | Execution | Verify branch/worktree intent before staging, committing, or pushing | | change-reaudit | Execution | Re-audit code changes for side effects, regressions, and edge cases | | cjk-text-wrap-audit | Execution | Diagnose CJK text-wrapping bugs with evidence-first cascade and verification checks | | competitive-feature-benchmark | Execution | Compare competing products at the UX and interaction level | | docs-finalize-and-commit | Execution | Finalize documentation with convention discovery and consistency checks | | finalize-and-commit | Execution | Finalize code for production — dedup, hardcode audit, clean commits | | frontend-decision-sync | Execution | Persist reusable frontend decisions into project documentation | | gh-ship-pr | Execution | Push a committed branch, register a ready PR, optionally wait for CI, address review comments, and merge | | notion-format | Execution | Auto-detect content type and format rich Notion documents | | oss-code-analysis | Execution | Analyze OSS repos at the code level (compare or deep-dive mode) | | ux-sentinel | System | Auto-detect recurring UX concepts and persist to Notion DB |

Install a skill

Install all skills at once from the skills.sh marketplace:

npx skills add comsky/remy-skill-recipes

Or copy a single skill manually:

# Cursor
cp -r skills/change-reaudit ~/.cursor/skills/
Claude Code
cp -r skills/change-reaudit ~/.claude/skills/
VS Code + Copilot (auto-discovers .claude/skills/ or use dedicated path)
cp -r skills/change-reaudit ~/.copilot/skills/

The agent will automatically discover and activate the skill when a matching task appears.

Use directly

• Identify whether your task needs:

• Read the Inputs Required section carefully.
• Provide complete context.
• Validate output using the skill's checklist.

Skill Format

Every skill is a folder containing a SKILL.md file with YAML frontmatter:

---
name: change-reaudit
description: >
  Re-audit code changes to identify side effects, regression risks,
  and unhandled edge cases before merging or deploying.
license: MIT
compatibility:
  - Claude Code
  - Cursor
metadata:
  type: execution
  category: review
  maturity: stable
  estimated_time: 10 min
---

Agents read only name and description during discovery (~100 tokens). The full markdown body loads on activation (<5000 tokens).

Skill Types

1. Execution Skills

Single-run structured workflows.

Used for:

• Code review
• Change auditing
• Competitive benchmarking
• Documentation writing
• Refactoring validation

• Require explicit inputs
• Produce structured outputs
• Contain guardrails and failure patterns
• Include realistic examples

2. System Skills

Persistent or automation-oriented behaviors.

Used for:

• Continuous concept detection
• Knowledge tracking
• External DB synchronization
• Conversation-wide logic

• Define activation rules
• Maintain state (conversation or external DB)
• Specify side effects explicitly
• Include operational guardrails

Skill Categories

| Category | Description | | ------------- | ---------------------------------- | | review | Change audits, regression analysis | | research | Competitive feature analysis | | cleanup | Refactor and commit structuring | | documentation | Architecture specs, design docs | | automation | Persistent or DB-connected skills |

Skill Structure

Execution Skill Structure

• Purpose
• When to Use
• When NOT to Use
• Inputs Required
• Output Format
• Procedure
• Guardrails
• Failure Patterns
• Examples (minimum 2)

System Skill Structure

• Purpose
• Scope (triggers / non-triggers)
• Inputs / Signals
• Core Behavior (Detection → Decision → Action)
• Output / Side Effects
• Guardrails
• Failure Patterns
• Examples

Skill Maturity Levels

• Draft — experimental
• Stable — reliable for repeated use
• Production — validated in real workflows

Model Assumptions

Skills are model-agnostic.

Assume:

• A reasoning-capable LLM
• Structured context input
• Long-context support preferred

Philosophy

LLMs are unreliable without structure.

Structure reduces:

• Hallucination
• Ambiguity
• Context loss
• Overconfidence errors

System skills extend this philosophy into persistent decision-memory patterns.

Contribution

This repository is primarily maintained for personal reuse, but high-quality pull requests are welcome.

Rules:

• Use the correct template (template/execution-template.md or template/system-template.md)
• Follow the SKILL.md standard (YAML frontmatter required)
• Include realistic examples
• Document guardrails and failure patterns

License

MIT