Skip to content

squirrel289/pax

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PAX: Agent Skills Library

A greenfield, modular library of composable, general-purpose Agent Skills for LLM-driven automation and workflow orchestration.

Vision

Build reusable, parameterized skills that can be easily composed into higher-level workflows. All skills are designed for LLM agent invocation, enabling sophisticated automation through simple composition.

Quick Start

Basic Usage

Skills can be invoked directly or composed into workflows:

# Direct skill invocation
"Analyze this codebase using parallel-execution"

# Workflow composition
"Process PR #42 end-to-end" (uses process-pr workflow)

Skill Categories

  • Execution: Control flow and task orchestration
  • Tools: Concrete integrations (GitHub, APIs, etc.)
  • Interaction: Human-agent collaboration patterns
  • Workflow: Composed, domain-specific automations

Core Principles

1. Composability

All skills are atomic or composable, enabling flexible workflow construction.

workflow: process-pr
  ├─ parallel-execution (assess PR readiness)
  ├─ resolve-pr-comments (address feedback)
  └─ merge-pr (safe merge execution)

2. Parameterization

Skills accept parameters for maximum reuse across contexts.

parallel-execution:
  - tasks: [task1, task2, task3]
  - execution-mode: concurrent

yolo:
  - confidence-threshold: 0.7
  - max-retries: 3

3. Modularity

General skills (parallel-execution, tool invocation) separate from workflow-specific skills (process-pr).

4. Discoverability

Each skill includes comprehensive SKILL.md with parameters, usage examples, and composition patterns.

5. LLM-Optimized

All skills designed for natural language invocation and chaining by LLM agents.

Library Structure

pax/
├── docs/                          # Documentation
│   ├── README.md                  # This file
│   ├── GETTING_STARTED.md         # Quick start guide
│   ├── SKILL_COMPOSITION.md       # Composition patterns
│   └── EXAMPLES.md                # Real-world examples
│
├── skills/                        # All skills
│   ├── execution/                 # Execution patterns
│   │   ├── parallel-execution/
│   │   │   └── SKILL.md
│   │   └── sequential-execution/
│   │       └── SKILL.md
│   │
│   ├── tools/                     # Tool integrations
│   │   └── pull-request-tool/
│   │       └── SKILL.md
│   │
│   ├── interaction/               # Agent-human interaction
│   │   ├── yolo/
│   │   │   └── SKILL.md
│   │   └── collaborative/
│   │       └── SKILL.md
│   │
│   └── workflow/                  # Composed workflows
│       ├── resolve-pr-comments/
│       │   └── SKILL.md
│       ├── merge-pr/
│       │   └── SKILL.md
│       └── process-pr/
│           └── SKILL.md
│
└── SKILL_LIBRARY_PLAN.md         # Original plan

Available Skills

Execution Skills

Execute multiple independent tasks simultaneously for maximum efficiency.

When to use: Multiple analyses, independent file processing, concurrent reviews

Example:

"Analyze security, performance, and testing in parallel"

Execute dependent tasks in order where each task relies on previous results.

When to use: Build pipelines, ordered workflows, dependent steps

Example:

"Process PR: fetch → test → review → merge"

Tool Skills

Unified entry point for all GitHub pull request and issue management. Automatically selects the best backend (Copilot API or CLI) for the environment.

When to use: Any PR/issue management, review, comment, or merge operation—always start here for maximum compatibility and future-proofing.

Note: You may also invoke as pr-tool (see SKILL.md for details).

Example:

"Merge PR #42 using pull-request-tool"

Manage GitHub pull requests and issues using Copilot's built-in PR/issue APIs. Provides structured, agent-native operations for review, comment, resolve, and merge workflows.

When to use: In Copilot agent environments (VS Code, Copilot CLI, etc.) for maximum integration and reliability.

Example:

"Fetch PR #42 details using copilot-pull-request"

Interact with GitHub pull requests for review, comment management, and merge operations via the GitHub CLI. Used as a backend by the wrapper when Copilot APIs are unavailable.

When to use: CLI environments, or as a fallback for PR/issue management.

Example:

"Get all unresolved comments on PR #42"

Interaction Skills

Autonomous "just do it" mode - execute actions without confirmation.

When to use: Well-defined workflows, time-critical operations, low-risk automation

Example:

"Process all approved PRs in YOLO mode"

Human-in-the-loop interaction with confirmations and feedback.

When to use: High-risk operations, ambiguous requirements, learning scenarios

Example:

"Let's review and merge PR #42 together"

Workflow Skills

Systematically address and resolve pull request review comments.

Composes: pull-request-tool + sequential-execution + (yolo OR collaborative)

Example:

"Resolve all PR comments on #42"

Safely merge pull requests after comprehensive verification.

Composes: pull-request-tool + sequential-execution + (yolo OR collaborative)

Example:

"Merge PR #42 after verifying all checks pass"

End-to-end PR processing from review to merge.

Composes: parallel-execution + sequential-execution + pull-request-tool + resolve-pr-comments + merge-pr + (yolo OR collaborative)

Example:

"Process PR #42 from start to finish"

Composition Patterns

Pattern 1: Linear Workflow

process-pr:
  Step 1: Assess (parallel)
  Step 2: Verify (sequential)
  Step 3: Resolve (sequential)
  Step 4: Merge (sequential)

Pattern 2: Parallel Analysis

codebase-review:
  Parallel:
    - Security analysis
    - Performance analysis
    - Test coverage analysis
  Then:
    - Synthesize findings

Pattern 3: Mode Selection

Any workflow:
  IF user wants autonomous → use yolo
  IF user wants interactive → use collaborative

Pattern 4: Tool Orchestration

pull-request-tool (tool):
  Canonical entry point for all PR/issue management
  Delegates to:
    - copilot-pull-request (API backend)
    - gh-pr-review (CLI backend)
  Used by:
    - resolve-pr-comments
    - merge-pr
    - process-pr

Real-World Examples

Example 1: Automated PR Workflow

User: "Process PR #42 in YOLO mode"

Execution:
- process-pr skill
- Mode: yolo
- Steps:
  1. Parallel assessment (PR status + reviews + CI)
  2. Run local tests
  3. Resolve 3 comments automatically
  4. Wait for CI to pass
  5. Verify approvals
  6. Merge with squash
  7. Delete branch

Output: "PR #42 merged successfully"

Example 2: Batch PR Processing

User: "Process all approved PRs"

Execution:
- List approved PRs: #41, #42, #43
- parallel-execution spawns:
  - process-pr(#41, yolo)
  - process-pr(#42, yolo)
  - process-pr(#43, yolo)
- Each runs independently
- Report: "3 PRs processed and merged"

Example 3: Collaborative Review

User: "Help me review and merge PR #42"

Execution:
- process-pr skill
- Mode: collaborative
- Interactive steps with user approval at each stage
- User makes decisions on ambiguous items
- Agent executes approved actions

Extension Guide

Adding New Skills

  1. Choose category: execution, tools, interaction, or workflow
  2. Create directory: skills/<category>/<skill-name>/
  3. Write SKILL.md: Follow existing format
  4. Document:
    • Purpose and when to use
    • Parameters
    • Usage examples
    • Composition patterns
  5. Test: Verify skill works in isolation and composition

Skill Template

---
name: skill-name
description: One-line description
category: execution|tools|interaction|workflow
license: MIT
composed-from: [optional list of skills this composes]
---

# Skill Name

Description and purpose.

## When to Use

Specific scenarios...

## Parameters

Required and optional parameters...

## Usage Examples

Concrete examples...

## Quick Reference

Cheat sheet...

Best Practices

  1. Start with general skills: Use execution and tool skills
  2. Compose into workflows: Build domain-specific workflows
  3. Choose interaction mode: yolo for automation, collaborative for oversight
  4. Leverage parallelism: Use parallel-execution when tasks independent
  5. Document decisions: Explain choices in skill outputs
  6. Handle errors gracefully: Implement proper error recovery
  7. Test composition: Verify skills work together correctly

Performance Tips

  • Use parallel-execution for independent tasks: 3-10x faster
  • Use yolo mode for well-defined workflows: No confirmation delays
  • Use sequential-execution only when dependencies exist
  • Batch process multiple items with parallel-execution

Contributing

  1. Fork the repository
  2. Create skill in appropriate category
  3. Follow SKILL.md template
  4. Add to this README
  5. Submit PR with examples

License

MIT License - See individual skills for details

Support

Roadmap

  • Core execution skills (parallel, sequential)
  • GitHub PR tools
  • Interaction modes (yolo, collaborative)
  • PR workflows (resolve, merge, process)
  • Additional tool integrations (Jira, Slack, etc.)
  • More workflow patterns
  • Skill discovery and search
  • Validation and testing framework

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published