CLAUDE.md Pre-loading Options Analysis
Analysis of options for pre-loading cross-referenced files into CLAUDE.md
Output
CLAUDE.md Pre-loading Options
Current File Sizes
| File | Size | ~Tokens |
|---|---|---|
| CLAUDE.md (current) | 24KB | ~6K |
| Cross-referenced files: | ||
| gvAgentDesign.md | 159KB | ~40K |
| gvShellDesign.md | 87KB | ~22K |
| gvBUILD.md | 66KB | ~16K |
| konui/CLAUDE.md | 41KB | ~10K |
| gvCodingStandards.md | 36KB | ~9K |
| gvConnectionDesign.md | 25KB | ~6K |
| VIBE.md | 20KB | ~5K |
| konsole/CLAUDE.md | 19KB | ~5K |
| API.md | 17KB | ~4K |
| KONTASK-GUIDE.md | 12KB | ~3K |
| Service CLAUDE.md files (5) | 35KB | ~9K |
| TOTAL if all pre-loaded | 541KB | ~135K |
Option 1: Status Quo (Recommended)
Approach: Keep current CLAUDE.md as-is. Read files on-demand.
Cost per turn: 6K tokens base + ~2-10K for files read
Pros: Minimal baseline cost, only pay for what you use, always fresh content
Cons: Extra tool calls, ~1-2s latency per file read
Option 2: Tiered Pre-loading
Approach: Pre-load only the most frequently needed docs.
Suggested tier:
Pre-load (~17K tokens): ├── gvCodingStandards.md # Every turn needs this ├── VIBE.md # Coding philosophy └── KONTASK-GUIDE.md # Every turn creates kontasks Read on-demand: ├── gvBUILD.md # Only for gvShell work ├── gvShellDesign.md # Only for gvShell work ├── gvAgentDesign.md # Only for agent internals └── Service CLAUDE.md # Only when working on that service
Cost: ~23K tokens per turn (vs 6K now)
Option 3: Condensed Summary File
Approach: Create CLAUDE-SUMMARY.md with key points from all docs.
Target: ~15KB (3-4K tokens) covering essentials from all files
Pros: Single source, manageable size, always loaded
Cons: Maintenance burden (must update when source changes), loses detail
Option 4: Context-Aware Loading
Approach: Multiple CLAUDE.md files, loaded based on work area.
CLAUDE.md # Base (always loaded) CLAUDE-GVSHELL.md # Pre-load when --scope=gvshell CLAUDE-KONNECT.md # Pre-load when --scope=product
Cons: Claude Code doesn't support conditional loading today
My Recommendation
Stick with Option 1 (status quo) for now, with one enhancement:
- Add
gvCodingStandards.mdinline into CLAUDE.md (most frequently needed) - Keep other cross-references as links
- Result: ~15K tokens base instead of 6K, but saves reads on 80% of turns
Cost Analysis
| Scenario | Tokens/Turn | Cost @ $15/MTok |
|---|---|---|
| Current (6K base + avg 5K reads) | 11K | $0.17 |
| Tiered pre-load (23K) | 23K | $0.35 |
| Full pre-load (135K) | 135K | $2.03 |
Note: These are input token costs. Actual turn costs also include output tokens and tool calls.
Quick Actions
Original Request
No layout configured
Details
Raw Data
{
"id": "d000f2ce-1848-4109-8d53-4c21e15b681f",
"type": "general",
"status": "completed",
"title": "CLAUDE.md Pre-loading Options Analysis",
"description": "Analysis of options for pre-loading cross-referenced files into CLAUDE.md",
"context": {
"output": "<div style=\"background:#1e1e1e;color:#e0e0e0;padding:20px;border-radius:8px;font-family:'Consolas','Monaco',monospace;\"><h2 style=\"font-family:sans-serif;color:#fff;margin:0 0 16px 0;\">CLAUDE.md Pre-loading Options</h2><h3 style=\"font-family:sans-serif;color:#61afef;margin:16px 0 12px 0;\">Current File Sizes</h3><table style=\"width:100%;border-collapse:collapse;margin:12px 0;font-size:14px;\"><tr style=\"border-bottom:1px solid #3d3d3d;background:#2d2d2d;\"><th style=\"padding:8px 12px;text-align:left;color:#e5c07b;\">File</th><th style=\"padding:8px 12px;text-align:right;color:#e5c07b;\">Size</th><th style=\"padding:8px 12px;text-align:right;color:#e5c07b;\">~Tokens</th></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;color:#98c379;\">CLAUDE.md (current)</td><td style=\"padding:8px 12px;text-align:right;\">24KB</td><td style=\"padding:8px 12px;text-align:right;\">~6K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;background:#252525;\"><td style=\"padding:8px 12px;color:#7f848e;\" colspan=\"3\">Cross-referenced files:</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">gvAgentDesign.md</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">159KB</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">~40K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">gvShellDesign.md</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">87KB</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">~22K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">gvBUILD.md</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">66KB</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">~16K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">konui/CLAUDE.md</td><td style=\"padding:8px 12px;text-align:right;\">41KB</td><td style=\"padding:8px 12px;text-align:right;\">~10K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">gvCodingStandards.md</td><td style=\"padding:8px 12px;text-align:right;\">36KB</td><td style=\"padding:8px 12px;text-align:right;\">~9K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">gvConnectionDesign.md</td><td style=\"padding:8px 12px;text-align:right;\">25KB</td><td style=\"padding:8px 12px;text-align:right;\">~6K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">VIBE.md</td><td style=\"padding:8px 12px;text-align:right;\">20KB</td><td style=\"padding:8px 12px;text-align:right;\">~5K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">konsole/CLAUDE.md</td><td style=\"padding:8px 12px;text-align:right;\">19KB</td><td style=\"padding:8px 12px;text-align:right;\">~5K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">API.md</td><td style=\"padding:8px 12px;text-align:right;\">17KB</td><td style=\"padding:8px 12px;text-align:right;\">~4K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">KONTASK-GUIDE.md</td><td style=\"padding:8px 12px;text-align:right;\">12KB</td><td style=\"padding:8px 12px;text-align:right;\">~3K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">Service CLAUDE.md files (5)</td><td style=\"padding:8px 12px;text-align:right;\">35KB</td><td style=\"padding:8px 12px;text-align:right;\">~9K</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;background:#2d2d2d;\"><td style=\"padding:8px 12px;color:#e5c07b;font-weight:bold;\">TOTAL if all pre-loaded</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;font-weight:bold;\">541KB</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;font-weight:bold;\">~135K</td></tr></table><h3 style=\"font-family:sans-serif;color:#61afef;margin:24px 0 12px 0;\">Option 1: Status Quo (Recommended)</h3><div style=\"background:#2d2d2d;padding:16px;border-radius:6px;margin:12px 0;\"><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Approach:</strong> Keep current CLAUDE.md as-is. Read files on-demand.</p><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Cost per turn:</strong> 6K tokens base + ~2-10K for files read</p><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Pros:</strong> Minimal baseline cost, only pay for what you use, always fresh content</p><p style=\"margin:0;\"><strong style=\"color:#e06c75;\">Cons:</strong> Extra tool calls, ~1-2s latency per file read</p></div><h3 style=\"font-family:sans-serif;color:#61afef;margin:24px 0 12px 0;\">Option 2: Tiered Pre-loading</h3><div style=\"background:#2d2d2d;padding:16px;border-radius:6px;margin:12px 0;\"><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Approach:</strong> Pre-load only the most frequently needed docs.</p><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Suggested tier:</strong></p><pre style=\"background:#252525;padding:12px;border-radius:4px;margin:8px 0;overflow-x:auto;color:#e0e0e0;\">Pre-load (~17K tokens):\n├── <span style=\"color:#98c379;\">gvCodingStandards.md</span> <span style=\"color:#7f848e;\"># Every turn needs this</span>\n├── <span style=\"color:#98c379;\">VIBE.md</span> <span style=\"color:#7f848e;\"># Coding philosophy</span>\n└── <span style=\"color:#98c379;\">KONTASK-GUIDE.md</span> <span style=\"color:#7f848e;\"># Every turn creates kontasks</span>\n\nRead on-demand:\n├── <span style=\"color:#e5c07b;\">gvBUILD.md</span> <span style=\"color:#7f848e;\"># Only for gvShell work</span>\n├── <span style=\"color:#e5c07b;\">gvShellDesign.md</span> <span style=\"color:#7f848e;\"># Only for gvShell work</span>\n├── <span style=\"color:#e5c07b;\">gvAgentDesign.md</span> <span style=\"color:#7f848e;\"># Only for agent internals</span>\n└── <span style=\"color:#e5c07b;\">Service CLAUDE.md</span> <span style=\"color:#7f848e;\"># Only when working on that service</span></pre><p style=\"margin:8px 0 0 0;\"><strong style=\"color:#98c379;\">Cost:</strong> ~23K tokens per turn (vs 6K now)</p></div><h3 style=\"font-family:sans-serif;color:#61afef;margin:24px 0 12px 0;\">Option 3: Condensed Summary File</h3><div style=\"background:#2d2d2d;padding:16px;border-radius:6px;margin:12px 0;\"><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Approach:</strong> Create <code style=\"background:#252525;padding:2px 6px;border-radius:4px;\">CLAUDE-SUMMARY.md</code> with key points from all docs.</p><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Target:</strong> ~15KB (3-4K tokens) covering essentials from all files</p><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Pros:</strong> Single source, manageable size, always loaded</p><p style=\"margin:0;\"><strong style=\"color:#e06c75;\">Cons:</strong> Maintenance burden (must update when source changes), loses detail</p></div><h3 style=\"font-family:sans-serif;color:#61afef;margin:24px 0 12px 0;\">Option 4: Context-Aware Loading</h3><div style=\"background:#2d2d2d;padding:16px;border-radius:6px;margin:12px 0;\"><p style=\"margin:0 0 8px 0;\"><strong style=\"color:#98c379;\">Approach:</strong> Multiple CLAUDE.md files, loaded based on work area.</p><pre style=\"background:#252525;padding:12px;border-radius:4px;margin:8px 0;overflow-x:auto;color:#e0e0e0;\">CLAUDE.md <span style=\"color:#7f848e;\"># Base (always loaded)</span>\nCLAUDE-GVSHELL.md <span style=\"color:#7f848e;\"># Pre-load when --scope=gvshell</span>\nCLAUDE-KONNECT.md <span style=\"color:#7f848e;\"># Pre-load when --scope=product</span></pre><p style=\"margin:8px 0 0 0;\"><strong style=\"color:#e06c75;\">Cons:</strong> Claude Code doesn't support conditional loading today</p></div><h3 style=\"font-family:sans-serif;color:#61afef;margin:24px 0 12px 0;\">My Recommendation</h3><div style=\"background:#1a3a1a;border:1px solid #98c379;padding:16px;border-radius:6px;margin:12px 0;\"><p style=\"margin:0 0 12px 0;\"><strong style=\"color:#98c379;\">Stick with Option 1</strong> (status quo) for now, with one enhancement:</p><ul style=\"margin:0;padding-left:20px;\"><li style=\"margin-bottom:8px;\">Add <code style=\"background:#252525;padding:2px 6px;border-radius:4px;\">gvCodingStandards.md</code> inline into CLAUDE.md (most frequently needed)</li><li style=\"margin-bottom:8px;\">Keep other cross-references as links</li><li>Result: ~15K tokens base instead of 6K, but saves reads on 80% of turns</li></ul></div><h3 style=\"font-family:sans-serif;color:#61afef;margin:24px 0 12px 0;\">Cost Analysis</h3><table style=\"width:100%;border-collapse:collapse;margin:12px 0;font-size:14px;\"><tr style=\"border-bottom:1px solid #3d3d3d;background:#2d2d2d;\"><th style=\"padding:8px 12px;text-align:left;color:#e5c07b;\">Scenario</th><th style=\"padding:8px 12px;text-align:right;color:#e5c07b;\">Tokens/Turn</th><th style=\"padding:8px 12px;text-align:right;color:#e5c07b;\">Cost @ $15/MTok</th></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">Current (6K base + avg 5K reads)</td><td style=\"padding:8px 12px;text-align:right;\">11K</td><td style=\"padding:8px 12px;text-align:right;\">$0.17</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">Tiered pre-load (23K)</td><td style=\"padding:8px 12px;text-align:right;\">23K</td><td style=\"padding:8px 12px;text-align:right;\">$0.35</td></tr><tr style=\"border-bottom:1px solid #3d3d3d;\"><td style=\"padding:8px 12px;\">Full pre-load (135K)</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">135K</td><td style=\"padding:8px 12px;text-align:right;color:#e06c75;\">$2.03</td></tr></table><p style=\"color:#7f848e;margin-top:16px;font-style:italic;\">Note: These are input token costs. Actual turn costs also include output tokens and tool calls.</p></div>",
"requestedAt": "2026-01-09T00:00:00.000Z",
"requestId": "bc83e2c3-fee6-4b1d-b324-595178df11d0",
"choices": [
{
"label": "Keep status quo",
"value": "Let's keep the current on-demand loading approach. No changes needed.",
"primary": true
},
{
"label": "Inline gvCodingStandards",
"value": "Inline gvCodingStandards.md into CLAUDE.md since it's needed on almost every turn",
"primary": false
},
{
"label": "Create summary file",
"value": "Create a CLAUDE-SUMMARY.md with condensed key points from all cross-referenced docs",
"primary": false
},
{
"label": "Full tiered approach",
"value": "Implement the full tiered pre-loading approach with VIBE.md, gvCodingStandards.md, and KONTASK-GUIDE.md inlined",
"primary": false
}
],
"turnTiming": {
"totalMs": 89341,
"ttfeMs": 48,
"thinkingMs": 89293,
"toolExecutionMs": 0,
"toolCallCount": 0,
"thinkingPct": 100,
"toolsPct": 0,
"ttfePct": 0
}
},
"createdBy": "claude",
"createdAt": "2026-01-08T14:18:37.964Z",
"updatedAt": "2026-01-08T14:18:51.116Z",
"requestId": "bc83e2c3-fee6-4b1d-b324-595178df11d0",
"scope": "vibetools",
"tags": [
"claude-code",
"documentation",
"performance"
],
"targetUser": "claude"
}