Lock down Claude Code's unstable behavior
Disable 1M context, adaptive thinking, and auto-memory in settings — and pin the subagent model to Sonnet — for noticeably more reliable runs.
- Type: Prompt
- Tags: Coding Agent Hacks, Intermediate, Claude Code
- Updated: 2026-10-01
Prompt
Disable 1M context, adaptive thinking, and auto-memory in settings — and pin the subagent model to Sonnet — for noticeably more reliable runs. Many users have noticed quality dips driven by bloated context, override settings, and stale auto-memory burning tokens. Ex-Meta/Microsoft engineer Kun Chen shared a settings block that locks these down: hard-pin effort to high, kill the 1M-token mode, switch off adaptive thinking, and disable auto-memory. Restart Claude Code after saving. Implementation example: // ~/.claude/settings.json { "effortLevel": "high", "env": { "CLAUDE_CODE_DISABLE_1M_CONTEXT": "1", "CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING": "1", "CLAUDE_CODE_DISABLE_AUTO_MEMORY": "1", "CLAUDE_CODE_SUBAGENT_MODEL": "sonnet" } }
More prompts
- Expense Report Automation System — Save More Time
- Meeting Scheduling Link Standardization — Save More Time
- Invoice Generation and Delivery Automation — Save More Time
- Document Template Automation Library — Save More Time
- Calendar Management Automation Rules — Save More Time
- File Organization and Naming System — Save More Time
- Recurring Payment Automation Setup — Save More Time
- Data Entry Elimination Framework — Save More Time