Browse / General-purpose / claude-code-config
claude-code-config
arm3n
Claude Code research agent configuration — hooks, skills, memory, and MCP setup for Opus 4.6
Open arm3n/claude-code-config → Does not match the four rules.
Built for one agent
ships CLAUDE.md and nothing for other agents.
In its own words
A custom status line and hook system for Claude Code that surfaces real-time rate limit data, context window usage, and automated context management.
What it shows
Rate limit countdowns
The reset countdowns come directly from Anthropic's API response headers (anthropic-ratelimit-unified-5h-reset and anthropic-ratelimit-unified-7d-reset), not estimates. Countdowns format as Xm, XhYm, or XdYh depending on the time remaining.
How the probe works
A lightweight 1-token Haiku request is sent to api.anthropic.com every 5 minutes. The response headers contain utilization fractions and reset timestamps for each rate limit window. Results are cached to ~/.claude/plan-usage-cache.json so the status line renders instantly between probes.
Authentication uses the OAuth token from ~/.claude/.credentials.json (written by claude login).
Hooks
Context Guard (hooks/context-guard.js)
A PreToolUse hook that monitors context window usage via a cache file written by the status line on every refresh.
- 80% — warns Claude to start wrapping up
- 90% — tells Claude to stop and suggest
/handover
Uses additionalContext injection so it's advisory, not blocking.
Gemini Delegate (hooks/gemini-delegate.js)
A PreToolUse hook on Read operations that suggests delegating large file reads to Gemini MCP (1M context) when the file would consume 25% of remaining context. Never warns for files under 40KB.
Quoting 68% of the README. Read the rest on GitHub ↗
Read from the repository's git tree on 2026-08-17.