AI Issue Triage

Date: 2026-01-15 Status: ✅ Live workflow operational (uses Claude / api.anthropic.com per .github/workflows/issue-triage.yml)

⚠️ Doc partially stale (2026-05-26): most of the sections below describe the previous OpenAI-based testing harness, including a npm run test:workflows:real script and apps/web/src/__tests__/workflows/ai-issue-triage-{mock,real}.test.js files that were removed in #522. The live workflow itself still works (it calls Claude) — only the local-test surface around it is outdated. Sections that mention OPENAI_API_KEY, test:workflows:real, or ai-issue-triage-{mock,real} are no longer applicable. A full rewrite tracking the current Anthropic-based pipeline is tracked separately.

Quick Start
What It Does
Testing
Configuration Validation
Implementation Details

Quick Start

See AI Triage in Action (30 seconds)

bash

# Option 1: Mock tests (free, no setup needed)
npm run test:workflows:run -- ai-issue-triage-mock

# Option 2: Real OpenAI API (requires API key)
echo "OPENAI_API_KEY=sk-proj-YOUR_KEY_HERE" >> .env.local
npm run test:workflows:real

Expected Output:

✓ Bug Report: labels=['bug','help wanted'], priority='high', cost=$0.0001
✓ Feature: labels=['enhancement'], priority='medium'
✓ Docs: labels=['documentation'], priority='low'
✓ Question: labels=['question'], priority='low'

What It Does

Core Workflows

issue-triage.yml - Main AI Auto-Apply Workflow

Triggers: When ANY new issue is created
Process:
1. Sends issue title + body to OpenAI GPT-4o-mini
2. AI suggests 1-3 labels + priority level + reasoning
3. Validates labels against allowed list
4. Automatically applies labels to the issue
5. Posts comment explaining what was applied and why
Fallback: If API fails, posts warning and continues
Philosophy: Something is better than nothing - auto-apply immediately

apply-triage.yml - Legacy Apply Workflow (Backward Compatibility)

Triggers: When someone comments /apply-triage on an issue
Status: Kept for backward compatibility but mostly redundant now

User Experience

1. Create issue normally
   ↓
2. Wait ~30 seconds
   ↓
3. Labels AUTO-APPLIED by AI
   ↓
4. AI posts comment explaining reasoning
   ↓
5. Issue is ready for work!

Cost Analysis

Model: GPT-4o-mini

Volume	Cost per Issue	Monthly Cost
100 issues	$0.00006	~$0.01
500 issues	$0.00006	~$0.03
1000 issues	$0.00006	~$0.06

Total cost: Less than $1/month even at high volume

Testing

Test Commands (npm-based)

Command	Type	Cost	Speed	When to Use
`npm run test:workflows:run`	Unit + Mock	Free	<1s	Every commit
`npm run test:workflows:run -- ai-issue-triage-mock`	Mock only	Free	~400ms	Development
`npm run test:workflows:real`	Real API	~$0.0002	~8s	Before deploy
`npm run test:workflows`	All (with real tests in watch mode)	~$0.0002	~10s	Full validation

Path 1: Mock Tests (Free, Recommended for Development)

bash

npm run test:workflows:run -- ai-issue-triage-mock

Coverage:

✅ Bug report triage
✅ Feature request triage
✅ Documentation request triage
✅ Question triage
✅ Invalid label filtering
✅ API key errors
✅ Rate limit errors
✅ Network failures (timeout, DNS)
✅ Token usage tracking
✅ Cost estimation

Best for: Daily development, rapid iteration, no API key needed

Path 2: Real API Tests (Recommended for Local Dev Before Deploy)

bash

# 1. Get OpenAI API key
# Go to: https://platform.openai.com/api-keys
# Click: "Create new secret key"
# Copy: Your key (starts with sk-proj-...)

# 2. Add to .env.local
echo "OPENAI_API_KEY=sk-proj-YOUR_KEY_HERE" >> .env.local

# 3. Run real API tests
npm run test:workflows:real

# Or explicitly enable:
export RUN_REAL_API_TESTS=true
npm run test:workflows -- ai-issue-triage-real

Coverage:

✅ Real bug report with OpenAI
✅ Real feature request with OpenAI
✅ Real documentation request with OpenAI
✅ Real question with OpenAI

Cost: ~$0.0002 per run (~$0.01 per 50 runs)
Best for: Local validation before deployment

Path 3: Unit Tests Only (Free, Fast)

bash

npm run test:workflows:run

Coverage: 25 unit tests (pure logic validation)
Speed: <1ms
Best for: CI/CD, quick validation

Expected Output

Test Files  2 passed | 1 skipped (3)
     Tests  37 passed | 4 skipped (41)
   Duration  453ms

✅ 25 unit tests (pure logic, free)
✅ 12 mock integration tests (simulated API, free)
✅ 4 real API tests (actual OpenAI, ~$0.0002)

Troubleshooting Tests

Tests are skipped?

Real API tests skip if OPENAI_API_KEY is not set OR RUN_REAL_API_TESTS is not true
This prevents accidental API costs

"Invalid API key" error?

Verify key starts with sk-proj-
Check it's from https://platform.openai.com/api-keys
Ensure it hasn't expired
Verify it has available credits

Need verbose output?

bash

npm run test:workflows -- ai-issue-triage-mock --reporter=verbose

Run specific test?

bash

npm run test:workflows -- ai-issue-triage-mock -t "should triage a bug"

Watch mode (re-run on changes)?

bash

npm run test:workflows -- ai-issue-triage-mock --watch

Test Files

src/__tests__/workflows/
├── ai-issue-triage.test.js           # 25 unit tests
├── ai-issue-triage-mock.test.js      # 12 mock integration tests
└── ai-issue-triage-real.test.js      # 4 real API tests

Configuration Validation

Purpose

Keep triage categories in sync across:

.github/workflows/config/triage-categories.json (source of truth)
.github/workflows/issue-triage.yml (workflow references)
src/__tests__/workflows/*.test.js (test fixtures)
GitHub Projects V2 "Area" field (where labels are applied)

npm Validation Task

bash

npm run test:workflows:validate

What it checks:

✅ JSON syntax validity
✅ Required fields present (id, name, description)
✅ Categories in .yml files match config
✅ Categories in test fixtures match config
✅ GitHub Projects V2 "Area" field in sync (if GH_TOKEN set)

Local Validation (Without GitHub)

bash

npm run test:workflows:validate

Validates JSON and cross-references but skips GitHub sync check.

With GitHub Sync Check

bash

GH_TOKEN=$(gh auth token) npm run test:workflows:validate

Also validates against GitHub Projects V2 configuration.

Manual Validation (Shell Script)

bash

bash .github/workflows/validate-triage-config.sh

Flags:

--strict - Fail on any mismatches (for CI/CD)
--fix - Auto-sync where possible (limited by GitHub API)
--help - Show help

Validation Scenarios

Add New Category to Config

Edit .github/workflows/config/triage-categories.json
Add new category with id, name, description
Commit and push
Validation workflow runs automatically
If using GitHub Projects V2: Add option to "Area" field
Run: npm run test:workflows:validate
✅ All checks pass

Update Existing Category

Edit description in config
Run: npm run test:workflows:validate
✅ Passes (descriptions aren't validated against GitHub)

Fix Out-of-Sync Issues

Config has categories GitHub doesn't:

bash

# Add them to GitHub Projects V2 manually:
# 1. Go to Lantern App project > Area field settings
# 2. Add missing options
# 3. Re-run: npm run test:workflows:validate

GitHub has categories config doesn't:

bash

# Option A: Add to config
# Option B: Remove from GitHub Projects V2
# Run: npm run test:workflows:validate

GitHub Actions Workflow

File: .github/workflows/validate-triage-config.yml

Runs automatically on:

Pull requests that modify .github/workflows/config/triage-categories.json
Weekly schedule (Monday 9 AM UTC)
Manual trigger via Actions tab

Failure alert:

Posts Discord notification to #automation-alerts
Includes link to validation output
Suggests steps to fix

Implementation Details

Key Features & Safeguards

✅ Auto-Apply for ALL Issues

No bot detection - ALL issues get triaged (including bot-created ones)
Runs immediately on issue creation
Safe label application: Uses append-only --add-label (never deletes existing labels)
Philosophy: Something is better than nothing

✅ Label Validation

VALID_LABELS="bug,documentation,duplicate,enhancement,good first issue,help wanted,invalid,question,wontfix"

Only existing GitHub labels can be applied
AI cannot create new labels
Typos and hallucinations filtered out

✅ Permission Checks

PERMISSION=$(gh api repos/$REPO/collaborators/$COMMENTER/permission)
if [ "$PERMISSION" = "admin" ] || [ "$PERMISSION" = "write" ]; then
  # Allow apply-triage
fi

Only maintainers can use /apply-triage command
Contributors cannot auto-apply labels

✅ Graceful Fallback

API key missing → Warning comment posted
API error → Error comment posted with helpful message
Invalid JSON → Handled with sensible defaults
No issue creation is ever blocked

Setup for Production

1. Add OpenAI API Key

bash

# Get key from: https://platform.openai.com/api-keys
# Add to: Repository Settings → Secrets → Actions
Name: OPENAI_API_KEY
Value: sk-proj-...

2. Test Workflow

bash

# Create test issue manually
# Wait ~30 seconds
# Verify AI comment appears and labels are applied

3. Monitor Costs

bash

# Set budget alert in OpenAI dashboard
# Recommended: $10/month limit
# Check usage: https://platform.openai.com/usage

Cost Breakdown

Item	Cost	Frequency	Total/Month
Mock tests (local)	Free	Every commit	$0
Real API tests (local)	~$0.0002	Before deploy	<$0.01
Production triage (100 issues)	~$0.006	Monthly	<$0.01
Total	—	—	<$0.02

Files & Locations

Core Workflows:
.github/workflows/issue-triage.yml           # Main auto-apply workflow
.github/workflows/apply-triage.yml           # Legacy apply workflow
.github/workflows/validate-triage-config.yml # Validation workflow

Configuration:
.github/workflows/config/triage-categories.json  # Categories source of truth

Test Files:
src/__tests__/workflows/ai-issue-triage.test.js           # 25 unit tests
src/__tests__/workflows/ai-issue-triage-mock.test.js      # 12 mock tests
src/__tests__/workflows/ai-issue-triage-real.test.js      # 4 real API tests

Scripts:
scripts/validate-triage-consistency.js       # npm-based consistency checker

Documentation:
docs/engineering/github/workflows/AI_TRIAGE.md  # This file (single source of truth)

Testing Checklist (Before Production)

[ ] Add OPENAI_API_KEY to repository secrets
[ ] Run mock tests locally: npm run test:workflows:run -- ai-issue-triage-mock
[ ] Run real API tests: npm run test:workflows:real
[ ] Create test issue via GitHub UI (human user)
[ ] Verify AI comment appears and labels applied within 60 seconds
[ ] Check OpenAI usage dashboard: https://platform.openai.com/usage
[ ] Set up monthly budget alert ($10 recommended)
[ ] Test /apply-triage command as maintainer
[ ] Verify non-maintainers cannot use /apply-triage
[ ] Run validation: npm run test:workflows:validate

Rollback Plan

If needed, disable AI triage:

Option 1: Disable Workflows

bash

Actions → issue-triage.yml → ⋯ → Disable workflow
Actions → apply-triage.yml → ⋯ → Disable workflow

Option 2: Remove API Key

bash

Settings → Secrets → OPENAI_API_KEY → Delete
# Workflows will gracefully degrade with warning comments

Option 3: Delete Workflows

bash

git rm .github/workflows/issue-triage.yml
git rm .github/workflows/apply-triage.yml
git commit -m "Remove AI triage workflows"

Success Metrics

After deployment, track:

Accuracy: % of AI suggestions accepted by maintainers
Coverage: % of issues triaged by AI vs. manual
Time Savings: Average time to first triage
Cost: Actual monthly API costs
Errors: Failed API calls, invalid suggestions

References

Configuration: triage-categories.json
OpenAI Docs: https://platform.openai.com/docs/api-reference
GitHub Projects V2: https://docs.github.com/en/issues/planning-and-tracking-with-projects
Label Safety: LABEL_SAFETY.md

Status: ✅ Production Ready
Breaking Changes: ❌ None
Backward Compatible: ✅ Yes
Reversible: ✅ Yes (disable workflows or remove API key)

AI Issue Triage ​

Table of Contents ​

Quick Start ​

See AI Triage in Action (30 seconds) ​

What It Does ​

Core Workflows ​

User Experience ​

Cost Analysis ​

Testing ​

Test Commands (npm-based) ​

Path 1: Mock Tests (Free, Recommended for Development) ​

Path 2: Real API Tests (Recommended for Local Dev Before Deploy) ​

Path 3: Unit Tests Only (Free, Fast) ​

Expected Output ​

Troubleshooting Tests ​

Test Files ​

Configuration Validation ​

Purpose ​

npm Validation Task ​

Local Validation (Without GitHub) ​

With GitHub Sync Check ​

Manual Validation (Shell Script) ​

Validation Scenarios ​

Add New Category to Config ​

Update Existing Category ​

Fix Out-of-Sync Issues ​

GitHub Actions Workflow ​

Implementation Details ​

Key Features & Safeguards ​

✅ Auto-Apply for ALL Issues ​

✅ Label Validation ​

✅ Permission Checks ​

✅ Graceful Fallback ​

Setup for Production ​

1. Add OpenAI API Key ​

2. Test Workflow ​

3. Monitor Costs ​

Cost Breakdown ​

Files & Locations ​

Testing Checklist (Before Production) ​

Rollback Plan ​

Success Metrics ​

References ​

AI Issue Triage

Table of Contents

Quick Start

See AI Triage in Action (30 seconds)

What It Does

Core Workflows

User Experience

Cost Analysis

Testing

Test Commands (npm-based)

Path 1: Mock Tests (Free, Recommended for Development)

Path 2: Real API Tests (Recommended for Local Dev Before Deploy)

Path 3: Unit Tests Only (Free, Fast)

Expected Output

Troubleshooting Tests

Test Files

Configuration Validation

Purpose

npm Validation Task

Local Validation (Without GitHub)

With GitHub Sync Check

Manual Validation (Shell Script)

Validation Scenarios

Add New Category to Config

Update Existing Category

Fix Out-of-Sync Issues

GitHub Actions Workflow

Implementation Details

Key Features & Safeguards

✅ Auto-Apply for ALL Issues

✅ Label Validation

✅ Permission Checks

✅ Graceful Fallback

Setup for Production

1. Add OpenAI API Key

2. Test Workflow

3. Monitor Costs

Cost Breakdown

Files & Locations

Testing Checklist (Before Production)

Rollback Plan

Success Metrics

References