Repo Scout — Continuous Learning from the Ecosystem

Trigger

• User says "scout repos", "find repos", "what's new on GitHub"
• Periodic execution via GHA or session start
• When discovering a new framework/tool and wanting to find usage examples

Process

1. Discovery (parallel sub-agents)

For each search query, launch a background agent:

gh search repos --sort=stars --limit=10 "QUERY"

Default search queries (expand as stack evolves):

• e2b sandbox agent — sandbox-based agents
• claude agent sdk — Claude Agent SDK usage
• mcp server tools — MCP integrations
• ai agent browser automation — browser agents
• composio agent — Composio-based agents
• vercel ai sdk chat — AI chat UIs
• assistant-ui chat — assistant-ui implementations
• shadcn agent — shadcn/ui + agent combos
• stagehand browser — Stagehand browser automation
• copilotkit agent — CopilotKit implementations

2. Analysis (per repo)

For each promising repo (>50 stars, recent activity):

gh api repos/{owner}/{name}/git/trees/main?recursive=1

Extract:

• Architecture: File structure, package.json deps, framework choices
• Agent patterns: How they handle tool calls, state machines, error recovery
• UI patterns: Status updates, split-screen, generative UI
• Browser patterns: E2B, Playwright, Stagehand, computer use
• Harness patterns: Skills, memory, context management

3. Document

Write findings to resources/read/repo-scout-{date}.md:

# Repo Scout — {date}

## {repo-name} ({stars} stars)

- **URL**: {url}
- **Stack**: {frameworks used}
- **Key pattern**: {one-line insight}
- **Files to study**: {list of key files}
- **Adoptable for Energy**: {what we can use}

4. Integrate

• Update resources/awesome-repos.md with new entries
• If a pattern is novel, create a learning in memory/LEARNINGS.md
• If a framework is new to us, add to MEMORY.md tech choices

Reference Repos (Permanent Watchlist)

Output

• resources/read/repo-scout-{date}.md — findings document
• Updated resources/awesome-repos.md — catalog
• New learnings in memory/LEARNINGS.md if applicable