Feature request: Increase MEMORY.md line limit (currently 200 lines)

Problem

With the 1M token context window now available in Claude Opus 4.6, the 200-line hard truncation of MEMORY.md is extremely restrictive for real-world multi-project setups.

• 200 lines ≈ 4KB of text
• 4KB is 0.004% of the available 1M token context
• Even 5,000 lines (~100KB) would be less than 1% of context

Use Case

I manage a 12-agent AI infrastructure across 39 VMs with 15+ active projects, 10+ blogs, email systems, and multiple client websites. The MEMORY.md index needs to reference:

• Agent configurations and IPs
• Active projects with links to detail files
• Credential references
• Operational rules and feedback
• Automated systems (crons, publishers, monitors)
• External server references

Even with aggressive compression (using MEMORY.md purely as an index with one-line entries pointing to individual files), I'm constantly at the 200-line limit and have to make painful tradeoffs about what to include.

Current Workaround

• Keep MEMORY.md as a minimal index (~80-120 lines)
• Store all actual content in individual .md files
• Reference via links that Claude reads on demand
• This works but means Claude starts every session without key context that would make it immediately productive

Proposal

• Increase the limit to 2,000-5,000 lines, or
• Make it configurable via settings.json (e.g., autoMemoryMaxLines: 2000), or
• Scale the limit based on the model's context window (e.g., 0.5% of max context)

The current 200-line limit made sense when context windows were 8K-32K tokens. With 200K-1M token models, it's an artificial constraint that significantly degrades the multi-session memory experience.

Environment

• Claude Code 2.1.x
• Model: Claude Opus 4.6 (1M context)
• OS: Linux (Debian 12)
• Use case: Enterprise infrastructure orchestration with persistent memory across sessions