AI-Powered Issue Operations

Capability: AI-assisted issue summarization, auto-labeling, Q&A, and documentation generation using gh-models

When to use: Leveraging LLMs for intelligent issue processing and automation

Installation Required: gh extension install github/gh-models

---

Quick Start

List Available Models

# Show all 29+ models
gh models list

# Popular models for issue operations:
# - openai/gpt-4.1
# - openai/gpt-4o-mini
# - anthropic/claude-3.5-sonnet

Basic Usage

# Run AI model
gh models run "openai/gpt-4.1" "Your prompt here"

# With multi-line prompt
gh models run "openai/gpt-4.1" "$(cat <<'EOF'
Analyze this issue and suggest improvements:
- Title clarity
- Completeness
- Priority assessment
EOF
)"

---

Common Workflows

1. Issue Summarization (88% effectiveness)

# Get issue content
ISSUE_BODY=$(gh issue view 123 --json body --jq .body)

# Summarize
gh models run "openai/gpt-4.1" "$(cat <<'EOF'
Summarize this issue in 2-3 bullet points:

$ISSUE_BODY
EOF
)"

Use Case: Creating concise summaries for long issues, weekly reports

---

2. Auto-Label Suggestion (89% effectiveness)

# Get issue content
ISSUE_CONTENT=$(gh issue view 123 --json title,body --jq '{title, body}')

# Available labels
LABELS="bug,feature,documentation,question,enhancement,wontfix,duplicate"

# Suggest labels
gh models run "openai/gpt-4.1" "$(cat <<'EOF'
Suggest 2-3 labels from this list: $LABELS

Issue:
$ISSUE_CONTENT

Respond with comma-separated label names only.
EOF
)"

# Apply suggested labels
gh issue edit 123 --add-label bug,priority:high

Use Case: Automating issue triage, maintaining consistent labeling

---

3. Issue Q&A (91% effectiveness)

# Knowledge base Q&A
QUERY="How do I use Claude Code plan mode?"

# Search relevant issues
ISSUES=$(gh search issues "$QUERY" --repo=terrylica/claude-code-skills-github-issues --json number,title,body --jq '.')

# Ask AI
gh models run "openai/gpt-4.1" "$(cat <<'EOF'
Answer this question based on these GitHub Issues:

Question: $QUERY

Issues:
$ISSUES

Provide a concise answer with issue references.
EOF
)"

Use Case: Knowledge base Q&A, finding relevant information across issues

---

4. Documentation Generation (86% effectiveness)

# Get related issues
ISSUES=$(gh search issues --label=feature-request --closed --json title,body --jq '.')

# Generate changelog
gh models run "openai/gpt-4.1" "$(cat <<'EOF'
Generate a user-facing changelog from these closed feature requests:

$ISSUES

Format:
## New Features
- Feature name: Brief description

Keep it concise and user-friendly.
EOF
)"

Use Case: Generating changelogs, release notes, feature documentation

---

5. Issue Classification

# Get issue
ISSUE=$(gh issue view 123 --json title,body --jq '{title, body}')

# Classify
gh models run "openai/gpt-4.1" "$(cat <<'EOF'
Classify this issue into ONE category:
- Bug Report
- Feature Request
- Documentation
- Question
- Enhancement

Issue:
$ISSUE

Respond with category name only.
EOF
)"

---

Effectiveness Metrics (Empirical Testing)

Operation	Effectiveness	Test Count
Issue Summarization	88%	5 tests
Auto-Label Suggestion	89%	5 tests
Issue Q&A	91%	5 tests
Documentation Generation	86%	5 tests
Issue Classification	88%	5 tests

Average Effectiveness: 88%

Detailed Results: GH-MODELS-POC-RESULTS.md

---

Model Selection

Fast & Cheap (Good for bulk operations):

• openai/gpt-4o-mini - Fast, cost-effective
• openai/gpt-3.5-turbo - Balanced

High Quality (Complex analysis):

• openai/gpt-4.1 - Best quality
• anthropic/claude-3.5-sonnet - Long context, detailed analysis

Testing: Try different models to find best quality/cost tradeoff

---

Best Practices

1. Test prompts first - Verify output quality before automation
2. Provide context - Include relevant labels, repo info in prompt
3. Be specific - Clear instructions = better results
4. Iterate - Refine prompts based on output quality
5. Validate output - AI can make mistakes, always verify
6. Rate limits - Be aware of API rate limits for batch operations

---

Limitations

• API rate limits - Check GitHub API limits for your account
• Cost - Some models have usage costs
• Accuracy - Not 100% reliable, human review recommended
• Context size - Very long issues may hit token limits
• No state - Each call is independent, no conversation memory

---

Integration Example: Auto-Triage Workflow

#!/bin/bash
# Auto-triage new issues

# Get new issues
gh issue list --label needs-triage --json number,title,body --jq '.[] | @json' | \
while read -r issue; do
  # Extract fields
  number=$(echo "$issue" | jq -r .number)
  content=$(echo "$issue" | jq -r '{title, body}')

  # Get AI suggestions
  labels=$(gh models run "openai/gpt-4o-mini" "$(cat <<EOF
Suggest 2-3 labels: bug,feature,documentation,question,enhancement
Issue: $content
Respond with comma-separated labels only.
EOF
  )")

  # Apply labels
  gh issue edit "$number" --add-label "$labels" --remove-label needs-triage

  echo "Triaged issue #$number: $labels"
done

---

Installation: gh extension install github/gh-models

Full Extension Guide: GITHUB_CLI_EXTENSIONS.md

Complete POC Results: GH-MODELS-POC-RESULTS.md