feat: add Claude agent for automated PR review and CI/CD remediation

.claude/README.md

@@ -0,0 +1,147 @@
+# Claude PR Review Remediator
+
+This directory contains Claude Code agents for automating PR review feedback remediation and CI/CD issue fixing.
+
+## Structure
+
+```text
+.claude/
+├── agents/
+│   └── pr-review-remediator.md    # Main PR review remediation agent
+├── commands/
+│   └── analyze-pr.md           # Slash command for quick PR analysis
+└── README.md                   # This file
+```
+
+## Usage
+
+### Method 1: Direct Agent Invocation
+
+```bash
+# Use the Task tool to invoke the agent
+/agent pr-review-remediator
+```
+
+Then provide the PR number when prompted.
+
+### Method 2: Slash Command
+
+```bash
+# Use the slash command with a PR number
+/analyze-pr 1440
+```
+
+### Method 3: Manual Invocation
+
+Simply ask Claude to analyze a PR:
+```
+"Please analyze PR #1440 for review feedback and failing checks"
+```
+
+The agent will automatically be invoked based on the task description.
+
+## Features
+
+### CodeRabbit Integration
+- Automatically finds and parses CodeRabbit review comments
+- **Prioritizes "🤖 Prompt for AI Agents" sections** over code diffs
+- Uses natural language prompts to understand the intent
+- Implements fixes based on understanding + project standards (not copying code)
+- Validates suggestions before applying them
+- Distinguishes between actionable items and nitpicks
+- Presents a validation analysis showing which suggestions are valid
+
+### GitHub Status Checks
+- Monitors all CI/CD checks (tests, linting, security scans)
+- Identifies root causes of failures
+- Provides targeted remediation strategies
+
+### Smart Linting
+- **ONLY lints changed files** in the PR (not the entire codebase)
+- Uses `make lint` which already implements `--new-from-rev=origin/main`
+- Targets fixes to specific files that were modified
+
+### Safety Controls
+- Always requires user approval before making changes
+- Validates all suggestions against project standards
+- Never modifies golden snapshots in tests/test-cases/
+- Creates feature branches for fixes (never pushes to main)
+
+## Example Workflow
+
+1. **Analyze PR**:
+   ```bash
+   /analyze-pr 1440
+   ```
+
+2. **Agent Response**:
+   - Fetches PR details and changed files
+   - Extracts CodeRabbit comments
+   - Checks CI/CD status
+   - Validates each suggestion
+   - Presents categorized action plan
+
+3. **User Approval**:
+   - Review the proposed fixes
+   - Approve valid suggestions
+   - Skip questionable ones
+
+4. **Execution**:
+   - Agent applies approved fixes
+   - Runs validation (lint, test, build)
+   - Prepares commit with descriptive message
+
+## Key Commands Used by Agent
+
+```bash
+# Get changed files in PR
+gh pr view <PR_NUMBER> --repo cloudposse/atmos --json files --jq '.files[].path'
+
+# Find CodeRabbit comments
+gh api repos/cloudposse/atmos/issues/<PR>/comments \
+  --jq '.[] | select(.user.login == "coderabbitai")'
+
+# Extract AI Agent prompts (PREFERRED)
+gh pr view <PR_NUMBER> --repo cloudposse/atmos --comments | \
+  grep -A 50 "Prompt for AI Agents"
+
+# Lint only changed files
+make lint  # Uses --new-from-rev=origin/main
+
+# Run tests for changed packages
+go test ./pkg/merge -v
+
+# Check PR status
+gh pr checks <PR_NUMBER> --repo cloudposse/atmos
+```
+
+## Validation Process
+
+The agent validates CodeRabbit suggestions by:
+
+1. **Reading AI Prompts First**: Extracts "🤖 Prompt for AI Agents" sections
+2. **Understanding Intent**: Analyzes the natural language explanation
+3. **Alignment with Standards**: Does it match CLAUDE.md requirements?
+4. **Real Issue vs Preference**: Is it fixing an actual problem?
+5. **Breaking Changes**: Will it break existing functionality?
+6. **Consistency**: Does it match existing codebase patterns?
+7. **Golden Snapshots**: Does it respect test fixtures?
+8. **Custom Implementation**: Writes fixes based on understanding, not copying
+
+Each suggestion is marked as:
+- ✅ **Valid**: Safe to apply
+- ⚠️ **Needs Review**: Requires discussion
+- ❌ **Skip**: Should not be applied
+
+## Customization
+
+To modify the agent behavior, edit:
+- `.claude/agents/pr-review-remediator.md` - Main agent logic
+- `.claude/commands/analyze-pr.md` - Slash command template
+
+## Best Practices
+
+1. Always review the agent's analysis before approving fixes
+2. Use the agent for routine feedback (formatting, linting, simple fixes)
+3. Handle complex architectural changes manually
+4. Keep the agent updated with project-specific requirements