Agents inappropriately report time estimates and knowledge requirements

Issue 1: Agents Report Time Estimates

Problem: Agents provide time estimates despite having no ability to predict actual implementation duration.

Examples:

• "This will take 2-3 days"
• "Migration effort: 1-2 days"
• "Effort: 3-5 days"

Why Wrong:

• Agents don't implement in real-time
• User works at their own pace
• Wastes ~50-100 tokens per response
• Creates false expectations

Fix: Report complexity (simple/moderate/complex), not duration.

---

Issue 2: Agents Report "Knowledge Required"

Problem: Agents warn users about needing to "learn X" when the agent itself implements the code.

Examples:

• "Requires learning Rust"
• "Need to know WASM"
• "High learning curve"

Why Embarrassing:

• Agents ARE the expert and write the code
• User doesn't need to learn anything
• Reveals confusion about agent role
• Wastes ~30-80 tokens per warning

Fix: Report technical tradeoffs (ecosystem maturity, debugging complexity), not user skill requirements.

---

Impact

Token Waste: ~100-200 tokens per technical response
User Experience: Requires manual editing of all outputs
Scale Issue: Every benchmark/analysis requires cleanup

---

Requested Fix

Update Claude Code system instructions:

1. ❌ Remove all time estimate examples and prompts
2. ❌ Remove all "user needs to learn X" patterns
3. ✅ Focus on actionable technical information only

---

Reproduction

Any Claude Code session where agents provide technical analysis or recommendations will exhibit both issues. Test with:

Prompt: "Compare React, Svelte, and Rust WASM for a dashboard"
Expected: Technical tradeoffs only
Actual: Time estimates + "requires learning Rust" warnings

---

User Workaround

Added to ~/.claude/CLAUDE.md:

## NEVER Report:
- Time estimates (always wrong)
- Knowledge requirements (agents write code, not users)

However, system instructions appear to override user config.