docs: Improve lead agent guidelines based on session learnings

[agent-relay]

Summary

Refactored lead agent guidelines to be more concise while incorporating practical insights from coordinating complex multi-agent sessions.

Key Improvements

1. Consolidated Core Principles (4→6)

   • Added emphasis on isolation (separate branches/PRs prevent chaos)
   • Added documentation strategy (trails/beads for future context)
   • Added communication cadence importance
   • Added key insight: "Investigate blockers deeply, but delegate the fix"

2. New Section: When to Spawn vs Assign

   • Clear guidance on when to spawn specialized agents
   • Blocker investigation strategy
   • Agent release process

3. Streamlined Workflow

   • Reduced from 6 detailed steps to 6 concise bullets
   • Removed redundant examples and duplicate patterns
   • More actionable, less prescriptive

4. Anti-Patterns with Examples

   • Simplified to concrete ❌→✅ comparisons
   • Easier to remember and apply

5. File Size Reduction

   • 179 lines → 120 lines (reduced by 33%)
   • Clearer focus without losing guidance

Learnings Incorporated

• Delegation effectiveness: Proper delegation enables parallel work and quality
• Documentation matters: Trails/beads help future agents understand context
• Isolation prevents chaos: Separate branches/PRs keep work clean and reviewable
• Blocker strategy: Investigate root cause, propose solution, spawn agent to implement
• Communication: Regular check-ins keep team aligned

Changes

• .claude/agents/lead.md: Consolidated and improved guidelines

[khaliqgant]

@willwashburn want to build out more of the these, but the most important right now IMO is the Lead.