Deployment Guide¶

Overview¶

aieng-bot can be used in two ways: 1. CLI - Fix any PR directly using aieng-bot fix --repo owner/repo --pr 123 2. GitHub Workflows - Automated monitoring and fixing of bot PRs (Dependabot, pre-commit-ci) across an organization

Deployment Checklist¶

Phase 1: Initial Setup¶

Create Anthropic API key from Anthropic Console
Create GitHub PAT with org-wide access
Add secrets to this repository:
ANTHROPIC_API_KEY
ORG_ACCESS_TOKEN
Enable GitHub Actions
Verify workflows appear in Actions tab

Phase 2: Testing¶

Test via CLI (recommended):

aieng-bot fix --repo owner/repo --pr 123

Or test via GitHub workflow:

gh workflow run fix-pr-agent.yml \
  --field target_repo="owner/repo" \
  --field pr_number="123"

Verify bot comments appear on test PR
Check workflow logs for errors

Phase 3: Monitoring¶

Enable scheduled runs (daily at 00:00 UTC)
Monitor for a few days
Review PRs the bot processes
Adjust skill templates if needed

Quick Deployment¶

# 1. Install aieng-bot
uv sync  # or pip install aieng-bot

# 2. Set environment variables
export ANTHROPIC_API_KEY="your-key"
export GITHUB_TOKEN="your-token"

# 3. Test on a PR
aieng-bot fix --repo owner/repo --pr 123

# 4. (Optional) For automated workflows, add secrets to GitHub:
# Settings → Secrets → Actions → Add ANTHROPIC_API_KEY and ORG_ACCESS_TOKEN

What Happens After Deployment¶

First Day¶

Discovery workflow runs at 00:00 UTC
Scans all VectorInstitute repos
Finds bot PRs (Dependabot and pre-commit-ci)
Classifies failures with Claude Haiku 4.5
Dispatches fix jobs for failing PRs
Auto-merges passing PRs

First Week¶

Bot handles most bot PRs automatically
Team sees reduced manual PR review load
Fix success rate becomes visible on Dashboard

Monitoring¶

Dashboard¶

View comprehensive analytics at platform.vectorinstitute.ai/aieng-bot: - PR status tracking - Agent execution traces - Success rates and metrics

CLI Commands¶

# Check recent runs
gh run list --workflow=discover-and-dispatch.yml --limit 10

# Check for failures
gh run list --workflow=discover-and-dispatch.yml --status failure --limit 5

# View specific run
gh run view RUN_ID --log

Success Metrics¶

Discovery workflow: 95%+ success rate
Auto-merge: 90%+ of passing PRs merged
Auto-fix: 50%+ of fixable issues resolved

Adjustments¶

Scan Frequency¶

Edit .github/workflows/discover-and-dispatch.yml:

on:
  schedule:
    - cron: '0 0 * * *'    # Daily (default)
    - cron: '0 */12 * * *' # Every 12 hours
    - cron: '0 */6 * * *'  # Every 6 hours

Exclude Repositories¶

Filter repos in workflow after getting the list:

REPOS=$(echo "$REPOS" | grep -v "repo-to-exclude")

Adjust Fix Behavior¶

Edit skill templates in .claude/skills/ to tune fix strategies.

Rollback Plan¶

Immediate Disable¶

gh workflow disable discover-and-dispatch.yml

Or via GitHub UI: Actions → Select workflow → "..." → Disable workflow

Full Rollback¶

Remove secrets from repository
Disable workflows
No cleanup needed in target repositories

Cost Estimation¶

Claude API¶

Classification: Claude Haiku 4.5 (low cost)
Fixing: Claude Sonnet 4.5 (moderate cost)
Estimated: $5-20/month depending on volume

GitHub¶

Actions minutes: Free for public repos, included in plan for private
API calls: Free within rate limits

Team Communication¶

Announcement Template¶

## 🤖 aieng-bot

We've deployed aieng-bot to help manage PR maintenance.

**What it does:**
- Fixes CI failures (linting, tests, security, build)
- Resolves merge conflicts
- Auto-merges PRs when all checks pass
- Comments on PRs it processes

**How to use:**
- CLI: `aieng-bot fix --repo owner/repo --pr 123`
- Automated workflows handle bot PRs (Dependabot, pre-commit-ci)
- Report issues at github.com/VectorInstitute/aieng-bot

Long-Term Maintenance¶

Monthly¶

Review bot effectiveness
Update skills based on patterns
Check API costs

Quarterly¶

Analyze metrics and trends
Review for model updates
Security and permissions audit

🤖 aieng-bot - AI-powered PR maintenance by Vector Institute AI Engineering