A rigorously structured collection of agent skills designed to reduce drift, enforce determinism, and make agentic work auditable. Built on the Open Agent-Skill Specification and extended with internal specifications that define canonical structures, execution patterns, and epistemic guardrails.

• Quickstart Guide — Getting started
• Skills Reference — Browse all available skills
• Skillsets — Orchestrator skills that coordinate member skills
• Contributing — Add your own skills

This repository treats language as inherently ambiguous and builds systems that succeed despite that ambiguity. Three principles guide all skill design:

• Determinism First, Subjectivity Last — Subjective reasoning must never compensate for missing deterministic outputs. If evidence is incomplete, return to earlier execution phases rather than explaining around gaps.

• Artifacts as Completion Signals — Success is mechanically verifiable. Skills declare what they produce; validation checks existence. Missing artifacts redirect execution to producing steps.

• Professional Skepticism — Agents interpret, clarify, and challenge rather than blindly comply. User input is valuable but not authoritative truth. Responsibility for correctness cannot be outsourced.

These principles are formalized in the Design Documentation, which includes epistemic tenets, interaction tenets, and a determinism framework.

All skills conform to the Open Agent-Skill Specification, ensuring portability across agent platforms. Each SKILL.md contains frontmatter-only definitions with explicit metadata for references, scripts, and keywords.

Beyond the open spec, this repository enforces internal specifications that define:

• SKILL.md Structure — Frontmatter schema, body constraints, validation rules
• References Structure — Canonical 7-file reference set (00_ROUTER through 06_FAILURES)
• Scripts Structure — Cross-platform script requirements and naming
• Skillset Structure — Orchestrator schema, pipelines, shared resources

These specifications ensure consistency across all skills and enable tooling to validate compliance.

All skills live in the skills/ directory. Browse them directly or use the reference documentation:

• Skills Reference — Complete list of skillsets and member skills with descriptions and keywords
• Skillsets Documentation — How orchestrator skills coordinate members, schema details, and usage patterns

Research-Plan-Implement (RPI):

• /rpi — Generic software development workflow
• /rpi-research — Codebase research and documentation
• /rpi-plan — Technical planning
• /rpi-plan-review — Expert review of implementation plans
• /rpi-plan-reimagine — Reimagine plans from scratch for optimization
• /rpi-implement — Plan execution
• /rpi-full — Full workflow with expert review and reimagine phases

Algorithm RPI:

• /algo-rpi — Algorithm development workflow
• /algo-rpi-research — Algorithm problem research
• /algo-rpi-plan — Algorithm planning with P0-P5 phases
• /algo-rpi-plan-review — AI/ML expert review of algorithm plans
• /algo-rpi-plan-reimagine — AI/ML expert reimagination for optimal design
• /algo-rpi-implement — Algorithm implementation with metrics
• /algo-rpi-full — Full workflow with AI expert review and optimization

Standalone Skills:

• /brainstorm — Interactive algorithm/AI solution brainstorming
• /code-review — Code quality review (scoped to modified files)
• /commit-message — Generate conventional commit messages
• /memory-bank — Manage persistent workspace context
• /doctor — Diagnostic protocol for complex failures

Reference-Only:

• coding-standards — Centralized code guidelines (accessed via .shared/ in other skills)

See the Quickstart Guide for installation, IDE integration, and first steps.

Utility scripts provide thin integration layers without duplicating skill content. Skills remain spec-compliant so they transfer cleanly when IDEs adopt agent skills natively.

Copy nunchuck skills to another project and generate workflows:

# Full deployment (copy skills + generate workflows)
bash scripts/deploy/deploy-to-project.sh /path/to/your/project

# Dry run to see what would be copied
bash scripts/deploy/deploy-to-project.sh /path/to/your/project --dry-run

# Copy skills only (generate workflows later)
bash scripts/deploy/deploy-to-project.sh /path/to/your/project --skills-only

This creates a .nunchuck/ directory in your target project (gitignored) containing skills and scripts, then generates workflows that reference them.

Generate IDE-specific configurations that reference skills without copying them:

# Windsurf: generates .windsurf/workflows/
bash scripts/adapters/windsurf/run.sh --skills-root skills --output-root .

# Cursor: generates .cursor/commands/
bash scripts/adapters/cursor/run.sh --skills-root skills --output-root .

Windows (PowerShell):

.\scripts\adapters\windsurf\run.ps1 --skills-root skills --output-root .
.\scripts\adapters\cursor\run.ps1 --skills-root skills --output-root .

Generate .SKILLS.md for agent discovery:

bash scripts/index/run.sh --skills-root skills

See CHANGELOG.md for release history.

See LICENSE for details.