@uniswap/ai-toolkit-nx-claude 0.5.16 → 0.5.18-next.0

package/dist/content/agents/agnostic/CLAUDE.md

@@ -0,0 +1,282 @@
+# AI Agents (Agnostic)
+
+## Purpose
+
+Language-agnostic AI agent definitions for Claude Code. Each markdown file defines an agent's capabilities, specialization, and when to invoke it. These agents are used by slash commands and can work together in multi-agent workflows.
+
+## Agent Files (32 total)
+
+### Code Quality Agents
+
+- `code-generator.md` - Generate production-ready code with tests
+- `code-explainer.md` - Explain code with deep analysis
+- `refactorer.md` - Advanced refactoring with safety checks
+- `style-enforcer.md` - Enforce code style and conventions
+- `security-analyzer.md` - Security vulnerability assessment
+- `performance-analyzer.md` - Performance analysis and optimization
+
+### Review & Analysis
+
+- `pr-reviewer.md` - Review PRs and address comments
+- `plan-reviewer.md` - Review implementation plans
+- `doc-writer.md` - Generate comprehensive documentation
+- `debug-assistant.md` - Advanced debugging with root cause analysis
+
+### Architecture & Design
+
+- `planner.md` - Create implementation plans
+- `researcher.md` - Conduct research (web + codebase)
+- `pattern-learner.md` - Learn and extract reusable patterns
+- `migration-assistant.md` - Guide version upgrades and migrations
+
+### Testing
+
+- `test-writer.md` - Generate comprehensive test suites
+- `test-runner.md` - Validate agent behaviors and test variations
+
+### Infrastructure & Operations
+
+- `infrastructure-agent.md` - Infrastructure automation and scaling
+- `cicd-agent.md` - CI/CD pipeline specialist
+- `deployment-engineer.md` - Configure deployments and CI/CD
+
+### Context & Documentation
+
+- `context-manager.md` - Manage context across agents
+- `context-loader.md` - Deep codebase understanding
+- `claude-docs-initializer.md` - Discover and create CLAUDE.md files
+- `claude-docs-manager.md` - Update CLAUDE.md based on changes
+- `claude-docs-fact-checker.md` - Verify CLAUDE.md accuracy
+
+### Git & Version Control
+
+- `pr-creator.md` - Create PRs with auto-generated messages
+- `commit-message-generator.md` - Generate structured commit messages
+- `stack-splitter.md` - Split Graphite PR stacks
+
+### Meta & Orchestration
+
+- `agent-orchestrator.md` - Coordinate multi-agent workflows
+- `agent-capability-analyst.md` - Analyze agent capabilities
+- `agent-optimizer.md` - Optimize agent performance
+- `prompt-engineer.md` - Analyze and optimize prompts
+- `feedback-collector.md` - Gather and analyze execution feedback
+- `claude-agent-discovery.md` - Discover and catalog agents
+
+## Agent Structure
+
+Each agent file follows a consistent markdown format:
+
+```markdown
+# Agent Name
+
+## Capabilities
+
+What this agent can do
+
+## Specialization
+
+Specific domain expertise
+
+## When to Invoke
+
+Conditions that warrant using this agent
+
+## Tools Available
+
+Tools this agent has access to
+
+## Output Format
+
+Expected output structure
+
+## Integration
+
+How it works with other agents
+```
+
+## Agent Categories
+
+### Proactive Agents
+
+Automatically invoked when their conditions are met:
+
+- `style-enforcer.md` - Auto-runs after code changes
+- `claude-docs-manager.md` - Auto-updates documentation
+- `security-analyzer.md` - Auto-scans for vulnerabilities
+
+### On-Demand Agents
+
+Explicitly invoked by commands or users:
+
+- `planner.md` - Called by `/plan` command
+- `researcher.md` - Called by `/research` command
+- `pr-reviewer.md` - Called by `/review-pr` command
+
+### Orchestration Agents
+
+Coordinate other agents:
+
+- `agent-orchestrator.md` - Routes tasks to specialists
+- `context-manager.md` - Manages shared context
+- `agent-optimizer.md` - Improves agent performance
+
+## Multi-Agent Workflows
+
+### Review Workflow
+
+```
+pr-reviewer → security-analyzer → performance-analyzer → doc-writer
+```
+
+### Refactoring Workflow
+
+```
+code-explainer → pattern-learner → refactorer → test-writer
+```
+
+### Planning Workflow
+
+```
+researcher → planner → plan-reviewer → context-manager
+```
+
+### Documentation Workflow
+
+```
+claude-docs-initializer → claude-docs-manager → claude-docs-fact-checker
+```
+
+## Development
+
+### Adding New Agents
+
+1. Create `new-agent.md` in this directory
+2. Follow the standard format (see existing files)
+3. Add to `index.ts` exports
+4. Update this CLAUDE.md file
+5. Test agent behavior in Claude Code
+
+### Modifying Existing Agents
+
+1. Edit the markdown file
+2. Test changes in multi-agent scenarios
+3. Update integration documentation
+4. Update this CLAUDE.md if capabilities change
+
+## Usage in Claude Code
+
+Agents are automatically discovered and loaded:
+
+```typescript
+import * as agents from '@ai-toolkit/agents-agnostic';
+```
+
+### Direct Invocation
+
+Via Task tool with subagent_type:
+
+```python
+Task(
+  subagent_type="planner",
+  prompt="Create plan for authentication feature"
+)
+```
+
+### Via Commands
+
+Commands invoke agents automatically:
+
+```bash
+/review-code  # Invokes multiple review agents
+/plan         # Invokes planner agent
+```
+
+### Agent Chaining
+
+Agents can spawn other agents:
+
+```markdown
+After analyzing, I'll invoke the refactorer agent...
+```
+
+## Agent Selection
+
+### Automatic Selection
+
+`agent-orchestrator` analyzes tasks and selects appropriate agents based on:
+
+- Task requirements
+- Agent capabilities
+- Current context
+- Performance history
+
+### Manual Selection
+
+Users/commands specify agents explicitly:
+
+```bash
+/review-code --agent=security-analyzer
+```
+
+## Best Practices
+
+### Agent Design
+
+- **Clear specialization**: Each agent has distinct expertise
+- **Tool access**: Request only necessary tools
+- **Output format**: Standardized, parseable outputs
+- **Error handling**: Graceful degradation
+- **Context awareness**: Use shared context efficiently
+
+### Multi-Agent Coordination
+
+- **Sequential**: When agents need each other's outputs
+- **Parallel**: When agents work independently
+- **Hierarchical**: Orchestrator → specialists
+- **Consensus**: Multiple agents vote on decisions
+
+### Performance
+
+- **Caching**: Reuse analysis across agents
+- **Batching**: Group similar operations
+- **Early exit**: Skip unnecessary agents
+- **Timeouts**: Prevent hanging workflows
+
+## Agent Communication
+
+### Shared Context
+
+Agents share data via `context-manager`:
+
+```typescript
+context.set('analysis', analysisResult);
+const prev = context.get('analysis');
+```
+
+### Agent Messages
+
+Agents can send structured messages:
+
+```json
+{
+  "from": "security-analyzer",
+  "to": "pr-reviewer",
+  "type": "finding",
+  "data": { "severity": "high", "issue": "..." }
+}
+```
+
+## Related Packages
+
+- `@ai-toolkit/commands-agnostic` - Commands that invoke agents
+- `@uniswap/ai-toolkit-nx-claude` - Nx integration and CLI
+- `@ai-toolkit/utils` - Shared utilities for agents
+
+## Orchestration Files
+
+- `index.ts` - TypeScript exports for agent registration
+
+## Auto-Update Instructions
+
+IMPORTANT: After changes to files in this directory, Claude Code MUST run `/update-claude-md` before presenting results to ensure this documentation stays synchronized with the codebase.

package/dist/content/agents/agnostic/pr-reviewer.md

@@ -1,7 +1,7 @@
 ---
 name: pr-reviewer
 description: Reviews PRs, addresses comments with plans or code changes, and fixes CI pipeline issues
-model: opus-4.1
+model: sonnet-4.5
 ---
 
 # pr-reviewer Agent

package/dist/content/commands/agnostic/CLAUDE.md

@@ -0,0 +1,206 @@
+# Slash Commands (Agnostic)
+
+## Purpose
+
+Language-agnostic slash command definitions for Claude Code. Each markdown file defines a command's behavior, parameters, and usage. These are the core commands available in the AI Toolkit.
+
+## Command Files (27 total)
+
+### Core Workflow Commands
+
+- `explore.md` - Deep dive into codebase areas
+- `plan.md` - Create implementation plans
+- `execute-plan.md` - Execute implementation plans step-by-step
+- `implement-spec.md` - Implement spec workflow tasks
+- `auto-spec.md` - Autonomous spec creation and implementation
+
+### Code Quality & Review
+
+- `review-code.md` - Multi-agent code review (architecture, security, performance)
+- `review-pr.md` - Comprehensive pull request review
+- `review-plan.md` - Review implementation plans
+- `fix-bug.md` - Diagnose and fix bugs with tests
+- `refactor.md` - Comprehensive refactoring with safety checks
+- `gen-tests.md` - Generate comprehensive test suites
+
+### Code Understanding
+
+- `explain-file.md` - Multi-agent code explanation
+- `research.md` - Research topics with web + codebase analysis
+
+### Git & PR Management
+
+- `create-pr.md` - Create/update Graphite PRs with auto-generated messages
+- `address-pr-issues.md` - Review and address PR comments/CI issues
+- `work-through-pr-comments.md` - Methodically work through PR comments
+- `generate-commit-message.md` - Generate structured commit messages
+- `split-stack.md` - Split Graphite PR stacks
+- `git-worktree-orchestrator.md` - Create and manage git worktrees with spec-workflow and Linear integration
+
+### Documentation
+
+- `claude-docs.md` - Initialize/update CLAUDE.md documentation
+- `claude-init-plus.md` - Discover and create CLAUDE.md files workspace-wide
+- `update-claude-md.md` - Update CLAUDE.md based on code changes
+
+### Development Operations
+
+- `deploy.md` - Orchestrate deployment pipelines
+- `monitor.md` - Set up application monitoring
+- `daily-standup.md` - Generate daily standup reports
+
+### Performance
+
+- `perf-analyze.md` - Performance analysis and optimization
+
+### Orchestration (Internal)
+
+- `index.ts` - TypeScript exports for command registration
+
+## File Structure
+
+Each command file follows a consistent markdown format:
+
+```markdown
+# Command Name
+
+## Overview
+
+Brief description of what the command does
+
+## Usage
+
+/command-name [arguments] [--flags]
+
+## Parameters
+
+- parameter1: description
+- parameter2: description
+
+## Behavior
+
+Detailed explanation of command behavior
+
+## Examples
+
+Example usage scenarios
+
+## Integration
+
+How it integrates with other commands/tools
+```
+
+## Command Categories
+
+### High-Complexity Commands (Multi-Agent)
+
+These commands orchestrate multiple specialized agents:
+
+- `/review-code` - Architecture, security, performance agents
+- `/review-pr` - Multiple review dimensions
+- `/explain-file` - Multiple analysis agents
+- `/refactor` - Safety and pattern agents
+- `/fix-bug` - Debugging and testing agents
+
+### Single-Purpose Commands
+
+Focused commands with clear, singular objectives:
+
+- `/generate-commit-message` - Git commit formatting
+- `/create-pr` - PR creation
+- `/perf-analyze` - Performance analysis
+
+### Workflow Commands
+
+Commands that manage larger processes:
+
+- `/auto-spec` - Full spec workflow
+- `/implement-spec` - Spec task implementation
+- `/execute-plan` - Plan execution
+- `/work-through-pr-comments` - Comment resolution workflow
+
+## Development
+
+### Adding New Commands
+
+1. Create `new-command.md` in this directory
+2. Follow the standard format (see existing files)
+3. Add to `index.ts` exports
+4. Update this CLAUDE.md file
+5. Test with `/new-command` in Claude Code
+
+### Modifying Existing Commands
+
+1. Edit the markdown file
+2. Test changes in Claude Code
+3. Update any affected documentation
+4. Update this CLAUDE.md if behavior changes
+
+## Usage in Claude Code
+
+Commands are automatically discovered and loaded:
+
+```typescript
+import * as commands from '@ai-toolkit/commands-agnostic';
+```
+
+Claude Code registers these as slash commands:
+
+```bash
+/explore <description>
+/plan <task description>
+/review-code [paths]
+```
+
+## Command Invocation
+
+### Direct Invocation
+
+User types command in Claude Code:
+
+```
+/review-code src/components/
+```
+
+### From Other Commands
+
+Commands can invoke other commands:
+
+```markdown
+After reviewing, run `/gen-tests` to add coverage
+```
+
+### From Agents
+
+Agents can recommend commands:
+
+```markdown
+I recommend running `/refactor src/utils/` to improve this code
+```
+
+## Best Practices
+
+### Command Design
+
+- **Single responsibility**: Each command does one thing well
+- **Clear parameters**: Document all arguments and flags
+- **Predictable behavior**: Same inputs = same outputs
+- **Composable**: Commands should work together
+- **Fail gracefully**: Handle errors with clear messages
+
+### Documentation
+
+- Keep markdown files up to date
+- Include examples for common use cases
+- Document edge cases and limitations
+- Link to related commands
+
+## Related Packages
+
+- `@ai-toolkit/commands-typescript` - TypeScript-specific implementations
+- `@ai-toolkit/agents-agnostic` - Agents used by commands
+- `@uniswap/ai-toolkit-nx-claude` - Nx integration and CLI
+
+## Auto-Update Instructions
+
+IMPORTANT: After changes to files in this directory, Claude Code MUST run `/update-claude-md` before presenting results to ensure this documentation stays synchronized with the codebase.