Implement Research Agent Phase 1: Literature Discovery Tool

[OnePunchMonk]

Summary

First phase of Research Agent roadmap: implement fetch_and_extract_content() tool for discovering strategy ideas from academic papers and industry research. This is the natural v7 capability building on verified harness foundation.

See docs/RESEARCH_AGENT_DESIGN.md for full architecture.

Related

• Blocks: Nothing (this is a new feature)
• Depends on: Make 6-epoch harness evolution results real & reproducible #19, Add CI job to verify harness evolution results don't become stale #20, Verify test suite: confirm 63 tests passing #21 (build on credible harness + verified tests)
• Relevant docs: docs/RESEARCH_AGENT_DESIGN.md, src/agent/tools/

Task

1. Implement fetch_and_extract_content() in src/agent/tools/registry.py:
   • Tool schema (JSON input spec)
   • Description for Claude
   • Execution logic
2. Wire into src/agent/tools/orchestrator.py for Claude tool-use integration
3. Test against 5-10 known finance URLs:
   • Academic papers (arXiv)
   • Industry research (papers.ssrn.com)
   • Strategy blogs
4. Verify output includes:
   • Extracted text content
   • Citation metadata (title, authors, date)
   • Relevance scoring

Files to Create/Modify

• src/agent/tools/registry.py — Add tool schema
• src/agent/tools/orchestrator.py — Wire into orchestrator
• tests/test_tools.py — Add integration tests
• tests/fixtures/test_urls.json — Known test URLs

Acceptance Criteria

• Tool loads and validates in verify_tools.py
• Claude can call it successfully via orchestrator
• Extracts text + metadata from real URLs
• 5+ test URLs return expected output format
• Tests added and passing
• Follows existing tool pattern (see registry.py for examples)

Notes

Start only after #19, #20, #21 are complete. This is the next real capability and builds on solid ground.