GitHub Autopilot + AI Reasoning Plane

Automated repo maintenance with AI-powered PR risk scoring and scout pattern extraction.

Architecture

github-autopilot.ps1  (Control Plane)        ai_repo_intel.py  (Reasoning Plane)
├── Repo discovery & ranking                  ├── PR risk scoring (Anthropic API)
├── Clone / update                            ├── Scout pattern extraction
├── Install / lint / test / build             ├── Report enrichment
├── GitHub Actions updates                    └── Heuristic fallback (no API needed)
├── PR enumeration & merge execution
├── Scout mode
└── Report generation (JSON + Markdown)

Quick Start

# One-time setup
Set-ExecutionPolicy -Scope Process Bypass
$env:ANTHROPIC_API_KEY = "sk-ant-..."

# Basic run
.\github-autopilot.ps1 -TopRepos 3 -CreateMaintenancePRs

# With AI scoring
.\github-autopilot.ps1 -TopRepos 3 -EnableAI -CreateMaintenancePRs

# Scout + AI
.\github-autopilot.ps1 -TopRepos 3 -ScoutMode -TopScoutRepos 5 -EnableAI

# Loop mode (runs every 6 hours)
.\github-autopilot.ps1 -TopRepos 3 -ScoutMode -Loop -LoopHours 6 -EnableAI

# Stricter merge threshold
.\github-autopilot.ps1 -TopRepos 5 -EnableAI -AIAutoMergeThreshold 15 -CreateMaintenancePRs

Parameters

Parameter	Default	Description
-Workspace	$HOME\gh-autopilot	Working directory
-TopRepos	5	Number of owned repos to process
-CreateMaintenancePRs	off	Enable PR creation and auto-merge
-ScoutMode	off	Clone starred repos for pattern analysis
-TopScoutRepos	5	Number of repos to scout
-Loop	off	Continuous execution mode
-LoopHours	6	Hours between loop cycles
-MaxPRLines	500	Skip PRs larger than this
-EnableAI	off	Activate AI reasoning plane
-AIPythonPath	python	Path to Python executable
-AIModulePath	(auto)	Path to ai_repo_intel.py
-AIAutoMergeThreshold	25	Max AI score for auto-merge

Hard Guardrails

• Skips draft PRs
• Skips non-mergeable PRs
• Skips PRs over 500 changed lines (configurable)
• Requires validation pass before auto-merge
• AI score must be <= threshold for auto-merge
• Separates owned and scouted workspaces
• Hard rules override AI judgment:
  • Failing tests = floor score 50
  • Security/deploy paths = cannot auto-merge (unless deps-only + tests pass)
  • Workflow changes = escalated to review
  • No-test repo + core logic = cannot auto-merge

AI Risk Scoring Bands

Score	Band	Action
0-25	auto_merge	Safe to merge
26-60	review	Needs human review
61-100	skip	Do not merge

Environment Variables

Variable	Required	Description
ANTHROPIC_API_KEY	For AI	Claude API key
AI_REPO_INTEL_MODEL	No	Default: claude-sonnet-4-20250514
AI_REPO_INTEL_DEBUG	No	Set "1" for verbose Python logs

Requirements

• PowerShell 5.1+ or 7+
• GitHub CLI (gh) authenticated
• Python 3.10+ (for AI features, stdlib only)
• Git