Implement GitHub Context Subagent

[prosdev]

Description

Implement a GitHub Context Subagent that indexes GitHub data (issues, PRs, discussions) and makes it semantically searchable. This extends dev-agent's context provision capabilities from code files to GitHub metadata, enabling AI assistants to have complete project context.

Core Mission Alignment

Goal: Provide relevant context to AI tools, reducing hallucinations

Current: We index code files and documentation
Missing: GitHub issues, PRs, discussions, comments
Solution: Index GitHub data like we index code

Acceptance Criteria

GitHub Data Indexing:

• Fetch issues via GitHub CLI (gh)
• Fetch PRs with metadata (commits, reviews, comments)
• Index issue/PR descriptions as vector embeddings
• Store GitHub metadata (labels, status, timestamps, links)
• Support incremental updates (only fetch changed data)

Semantic Search:

• Search issues: dev gh search "authentication bug"
• Search PRs: dev gh search --type pr "performance"
• Find related issues: dev gh related --issue 42
• Combined search (code + GitHub): dev search "oauth" --include-github

Context Provision:

• Get issue context: dev gh context --issue 42
• Returns: issue data, related PRs, linked code files, discussions
• Integrate with Planner (better issue understanding)
• Integrate with Explorer (find related issues to code patterns)

General:

• Integrates with subagent coordinator
• Uses existing vector storage (LanceDB)
• Respects .gitignore and privacy settings
• Handles rate limiting gracefully

Architecture

┌─────────────────────────────────────────────────────────┐
│                     dev-agent                            │
├─────────────────────────────────────────────────────────┤
│ Code Scanner    → Indexer → Vector Store (LanceDB)      │
│ GitHub Fetcher  → Indexer → Vector Store (LanceDB)      │
│                              ↓                           │
│                     Semantic Search                      │
│                    (code + GitHub)                       │
└─────────────────────────────────────────────────────────┘

Use Cases

1. Issue Context for AI

dev gh context --issue 42

# Returns full context:
# - Issue description
# - Related issues (#35, #28)
# - Related PRs (#40)
# - Affected code files (via links/mentions)
# - Discussion threads

2. Enhanced Planning

dev plan 42

# Planner now has access to:
# - Full issue description (not just gh CLI output)
# - Related issues for context
# - Previous similar work (from closed issues/PRs)

3. Pattern Discovery

dev explore pattern "error handling"

# Explorer finds:
# - Code patterns
# - Related issues discussing error handling
# - PRs that improved error handling

4. Knowledge Base

dev gh search "how do we handle rate limiting"

# Searches:
# - Issue discussions
# - PR descriptions
# - Code comments
# - Documentation

CLI Commands

# Indexing
dev gh index                     # Index all GitHub data
dev gh index --since 2024-01-01  # Incremental update
dev gh update                    # Refresh changed items

# Searching
dev gh search "query"            # Search all GitHub data
dev gh search "query" --type issue
dev gh search "query" --type pr

# Context
dev gh context --issue 42        # Get full context for issue
dev gh related --issue 42        # Find related issues/PRs

# Stats
dev gh stats                     # Show indexed data stats

Technical Implementation

Phase 1: GitHub Fetcher (Day 1)

• Use gh CLI to fetch issues and PRs
• Parse JSON output into structured types
• Handle pagination and rate limits
• Store raw data with metadata

Phase 2: Indexing (Day 1-2)

• Extract text from issues/PRs for embedding
• Generate vectors using Transformers.js (same as code)
• Store in LanceDB with GitHub-specific metadata
• Implement incremental updates

Phase 3: Search & Context (Day 2)

• Semantic search over GitHub data
• Context assembly (issue + related items)
• Integration with existing search

Phase 4: Agent Integration (Day 3)

• Expose via Subagent Coordinator
• Integrate with Planner agent
• Integrate with Explorer agent
• Message-based communication

Data Model

interface GitHubDocument {
  type: 'issue' | 'pr' | 'discussion';
  number: number;
  title: string;
  body: string;
  state: 'open' | 'closed';
  labels: string[];
  author: string;
  createdAt: string;
  updatedAt: string;
  comments: number;
  relatedIssues: number[];  // Extracted from links
  relatedPRs: number[];     // Extracted from links
  linkedFiles: string[];    // Mentioned in issue/PR
}

Dependencies

• Requires: gh CLI installed and authenticated
• Uses: Existing indexer and vector storage
• Integrates with: Planner (Implement Planner Subagent #8), Explorer (Implement Explorer Subagent #9), Coordinator (Implement Subagent Coordinator #7)

Success Metrics

• AI assistants can find relevant issues/PRs for any query
• Planner generates better task breakdowns with GitHub context
• Explorer discovers cross-cutting concerns (code + issues)
• Developers spend less time searching GitHub manually

Future Enhancements

• Index PR review comments
• Index discussion threads
• Track issue relationships (blocks/blocked-by)
• Temporal analysis (issue trends over time)
• Integration with GitHub Projects

Branch: feat/github-context-subagent
Priority: High (enables better AI assistance)
Estimate: 3 days
Parent Epic: #1 (Core Context Provider)