Skrills

Skills support engine for Claude Code, Codex CLI, and GitHub Copilot CLI.

Installation | User Guide | CLI Reference | MCP Tutorial | FAQ | Changelog

Features

Skrills validates skills against Claude Code (permissive), Codex CLI (strict), and Copilot CLI (strict) rules. It syncs skills, commands, agents, MCP servers, and preferences across all three environments, preventing configuration drift. The validation engine derives missing YAML frontmatter from file paths and content to fix common errors automatically.

For optimization, Skrills analyzes token usage per skill and suggests reductions to fit context windows. It resolves skill dependencies with cycle detection and semantic versioning constraints. The built-in MCP server provides over 40 tools for validation, sync, and project-aware skill generation, while session mining improves recommendations based on actual usage.

Demo

See the quickstart tutorial for a detailed walkthrough, or the MCP tutorial for server setup.

Installation

macOS / Linux:

curl -LsSf https://raw.githubusercontent.com/athola/skrills/HEAD/scripts/install.sh | sh

Windows PowerShell:

powershell -ExecutionPolicy Bypass -NoLogo -NoProfile -Command ^
"Remove-Item alias:curl -ErrorAction SilentlyContinue; iwr https://raw.githubusercontent.com/athola/skrills/HEAD/scripts/install.ps1 -UseBasicParsing | iex"

crates.io: cargo install skrills

See installation guide for HTTP transport setup, systemd services, and advanced options.

Quickstart

# Validate skills for Codex/Copilot compatibility
skrills validate --target all --autofix

# Analyze token usage
skrills analyze --min-tokens 1000 --suggestions

# Sync from Claude to all other CLIs
skrills sync-all

# Start MCP server
skrills serve

# Interactive mode
skrills tui

See CLI reference for all 30+ commands.

Why Skrills

Claude Code, Codex CLI, and Copilot CLI have different requirements for skill definitions. Codex and Copilot require YAML frontmatter with specific character limits (name max 100, description max 500), while Claude is permissive. Skrills catches these discrepancies at validation time, preventing runtime errors.

The sync system uses file hashing to respect manual edits, ensuring user changes aren't overwritten. Token analytics and dependency resolution help maintain a clean, efficient skill library within context limits.

Limitations

• No runtime skill injection: Skrills validates and syncs files; it does not inject skills into prompts at runtime
• Copilot command sync: Copilot CLI does not support slash commands, so command sync is skipped
• Empirical mining: Session history parsing works best with recent Claude Code / Codex CLI versions
• LLM generation: Requires ANTHROPIC_API_KEY or OPENAI_API_KEY for skill creation

Architecture

See architecture docs for details.

Configuration

Create ~/.skrills/config.toml for persistent settings:

[serve]
auth_token = "your-secret-token"
tls_auto = true
cors_origins = "https://app.example.com"

Precedence: CLI flags > environment variables > config file.

See security docs for TLS setup and FAQ for environment variables.

Documentation

Development

make lint test --quiet

Requires Rust 1.75+. See development guide for test coverage, CI, and contribution workflow.

Contributing

• Security issues: See security policy and threat model
• Bug reports: Include OS, skrills --version, and logs (--trace-wire for MCP)
• Pull requests: Follow process guidelines; update docs/tests with code

License

MIT