[FEATURE] Pull Requests as skill source — extract architectural guidance from PRs

[yusufkaraaslan]

🚀 Feature Description

Add PR (Pull Request) support as a skill source type, integrated into the existing GitHub scraper pipeline. PRs contain high-quality concentrated knowledge — diffs, descriptions, review comments, and linked issues — that can produce architectural guidance skills.

Integration with GitHub Source (not separate)

PRs are part of the GitHub source, not a standalone scraper. When using skill-seekers create owner/repo:

• PRs are included as part of the GitHub analysis (alongside code, docs, community streams)
• PR content enriches the codebase skill with architectural guidance
• User can opt to generate PR-only skills with a flag (e.g., --prs-only)

PR Collection with Optional Max Count

• Default: no cap — process all PRs (or all matching a filter)
• Optional --max-prs N to limit count
• Filter by label, author, date range, or state:

  skill-seekers create owner/repo --max-prs 50
  skill-seekers create owner/repo --pr-filter "label:architecture"
  skill-seekers create owner/repo --pr-filter "label:breaking-change"
  skill-seekers create owner/repo --prs-only --pr-filter "state:merged"

What to Extract from Each PR

1. Title and description — the "why" behind the change
2. Diff with context — what actually changed (summarized for large diffs)
3. Review comments — reviewer feedback, edge cases caught, design discussions
4. Linked issues — the problem being solved
5. Labels — categorization (bug, feature, breaking-change, etc.)
6. File change summary — which areas of the codebase were touched

Output: PR Skill + Architectural Guidance

• PR content produces skills focused on architectural decisions and patterns
• Supplements codebase analysis with "how and why the code evolved"
• When used with GitHub source: enriches the codebase skill's architecture section
• When used standalone (--prs-only): produces a focused "architectural guidance" skill

Freshness Warning

• PRs are point-in-time snapshots — code may have changed since
• Include metadata: PR date, merge status, whether files still exist
• Flag stale PRs (e.g., files deleted or heavily modified since merge)

Source Detection

• source_detector.py should recognize owner/repo#123 for single PR
• Collection mode activated by --pr-filter or --max-prs flags
• Default GitHub source includes PRs automatically (opt-out with --skip-prs)

Implementation Notes

• Extend github_scraper.py or github_fetcher.py to fetch PR data via GitHub API
• Use gh api repos/{owner}/{repo}/pulls for PR list and diffs
• Review comments via pulls/{number}/comments and pulls/{number}/reviews
• Large diffs should be summarized (configurable threshold)
• Add to the Three-Stream architecture: PRs are part of the Community stream

Example

# GitHub source with PRs included (default)
skill-seekers create owner/repo

# Limit PR count
skill-seekers create owner/repo --max-prs 100

# PR-only skill (architectural guidance)
skill-seekers create owner/repo --prs-only --pr-filter "label:architecture"

# Single PR (future: source detection)
skill-seekers create owner/repo#456