fixclaude

Fix Claude Code — Interactive Performance Wizard

Run an interactive wizard that detects the user's current Claude Code configuration, applies proven performance fixes, and optionally installs powerful companion tools. Walk through every step in plain language so any user — beginner or expert — understands what's happening and why.

Tone & Style

Keep explanations short, friendly, and jargon-free. One or two sentences per setting. Use simple analogies when helpful. Never assume the user knows what an environment variable is — just show what changes and why it helps.

Wizard Flow

Step 1: Welcome & Detect Current State

Display a brief welcome message explaining what the wizard does.

Read the user's current settings file at ~/.claude/settings.json. Parse the existing env block and top-level fields. Note which of the core fixes are already applied vs missing.

If the file doesn't exist, note that a fresh settings file will be created.

Step 2: Core Performance Fixes

Load references/core-settings.md for the detailed explanation of each setting.

Present all 8 core fixes in a clear table showing:

• Setting name
• What it does (plain English, one line)
• Current value → Recommended value

The 8 core fixes are:

Setting	Purpose
CLAUDE_CODE_EFFORT_LEVEL	Max reasoning every turn
CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING	Consistent deep thinking
MAX_THINKING_TOKENS	Maximum reasoning budget
CLAUDE_CODE_DISABLE_1M_CONTEXT	More compute for thinking
alwaysThinkingEnabled	Extended thinking every turn
CLAUDE_CODE_NO_FLICKER	Smooth screen rendering
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC	Disable telemetry and background network
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS	Multi-agent coordination

Note: Mention that CLAUDE_CODE_EFFORT_LEVEL=max uses more tokens per turn (higher cost on API plans; included on Max subscriptions). Also note that agent teams (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS) uses significantly more tokens when active (~7x) — it enables the capability but does not activate by default, so there is no cost unless the user explicitly uses agent teams.

After presenting the table, use AskUserQuestion to confirm: "Ready to apply these 8 performance fixes? (yes/no)"

On confirmation, apply settings to ~/.claude/settings.json. Read the file first, then use the Write tool to write the complete updated JSON (preserving all existing keys). Do NOT use the Edit tool for JSON — it is line-based and can corrupt JSON syntax. Instead, read the full file, merge the new keys in memory, and write the complete result.

After applying, show a short confirmation: what changed.

Step 3: Session Timer Bump (Rate Limit Optimizer)

After applying core fixes, offer to set up a persistent scheduled routine for rate limit optimization.

Use AskUserQuestion to ask: "Would you like to set up a session timer bump? Claude Code uses a rolling 5-hour usage window for rate limits. This creates a persistent cloud routine that pings every 3 hours to keep that window rolling, so you get more usage before hitting rate limits. Unlike session-only timers, this runs indefinitely — even when your computer is off. (yes/no)"

If no — skip to Step 4.

If yes — load references/session-timer.md for details. Since /schedule is a built-in interactive slash command, present the command to the user and ask them to run it:

Tell the user to paste this command:

/schedule

Then guide them through the interactive setup:

• Schedule: Every 3 hours
• Prompt/task: "Session timer bump — keeping the rolling usage window active. No action needed."

The /schedule command creates a cloud-hosted routine managed at claude.ai/code/routines. It runs on Anthropic's infrastructure, survives session restarts, works even when the user's computer is off, and never auto-expires.

After the user confirms the routine is created, explain:

• It runs every 3 hours on Anthropic's cloud — no need to keep Claude Code open
• They can view and manage it at claude.ai/code/routines or via /schedule list
• To remove it later: /schedule list to find it, then delete from the routines dashboard
• Daily run limits apply based on plan (Pro: 5/day, Max: 15/day, Team/Enterprise: 25/day)

Step 4: Optional Plugin Improvements

Use AskUserQuestion to ask: "Would you like to explore optional plugin improvements? These add a status dashboard, live docs, browser testing, and more. (yes/no)"

If no — skip to Step 5.

If yes — load references/optional-plugins.md and walk through each tool one at a time:

1. Claude HUD — Visual status bar showing context health, tokens, rate limits
2. Context7 MCP — Injects up-to-date library docs into your prompts
3. Playwright MCP — Claude can test its own UI changes in a real browser
4. Caveman skill — Shorter, more focused output
5. Claude in Chrome — Control your browser from the terminal

For each tool:

• Explain what it does in 1-2 plain sentences
• For MCP servers (Context7, Playwright): run the claude mcp add bash command directly
• For plugins (Claude HUD, Caveman): present the install command to the user and ask them to run it (these are interactive slash commands that cannot be executed via Bash)
• For Chrome: guide the user through browser extension setup and update ~/.claude.json (note: this is a different file from ~/.claude/settings.json)
• Confirm success or report any issues

Step 5: Summary

Load references/summary-template.md for the recap format.

Display a clear, organized summary of everything that was applied:

• Core settings applied (with before → after values)
• Plugins/tools installed (if any)
• Quick tips for getting the most out of the new setup

End with an encouraging message — the user's Claude Code is now optimized.

Important Notes

• Never overwrite existing settings. Merge changes into the existing file, preserving all hooks, permissions, plugins, and other configuration.
• Always read before writing. Read the current settings.json before making any edits.
• Handle missing files gracefully. If settings.json doesn't exist, create it with just the new settings.
• Be honest about trade-offs. Mention that max effort uses more tokens/costs more. The user should know.
• Skip what's already done. If a setting is already at the recommended value, note it as "already set" and move on.

Reference Files

For detailed information loaded on-demand:

• references/core-settings.md — Deep explanations, caveats, and sources for each core fix
• references/session-timer.md — Rate limit optimization via /schedule cloud routine — setup, management, and caveats
• references/optional-plugins.md — Install instructions and setup guides for each optional tool
• references/summary-template.md — Format template for the end-of-wizard recap