Feature: expose rate limit utilization in statusLine JSON input

Summary

The statusLine command receives a rich JSON payload via stdin (session info, model, cost, context window tokens, etc.), but it does not include rate limit utilization data. The /usage command displays "Current session" and "Current week (all models)" percentages — I'd like to show the same info in my custom status line.

Current behavior

The statusLine JSON input includes:

{
  "cost": { "total_cost_usd": ... },
  "context_window": {
    "used_percentage": 20,
    "remaining_percentage": 80,
    "total_input_tokens": ...,
    "total_output_tokens": ...,
    "current_usage": { ... }
  },
  "exceeds_200k_tokens": false
}

No rate limit / utilization fields are present.

Desired behavior

Add rate limit utilization to the statusLine JSON, mirroring what /usage shows:

{
  "rate_limit": {
    "five_hour": {
      "utilization": 0.42,
      "resets_at": 1741372800
    },
    "seven_day": {
      "utilization": 0.15,
      "resets_at": 1741737600
    },
    "seven_day_sonnet": {
      "utilization": 0.08,
      "resets_at": 1741737600
    },
    "status": "allowed"
  }
}

Fields can be null/absent when data hasn't been received yet (before the first API call) or when running with an API key that doesn't return unified rate limit headers.

Why

• The data already exists in memory (parsed from anthropic-ratelimit-unified-* response headers into the FG state variable)
• The statusLine builder function (cyq()) just needs to read and include it
• This lets users build status lines that show usage warnings without needing to open /usage manually
• Particularly useful for team/enterprise users monitoring session and weekly budgets

Workarounds considered

• Querying the API externally: OAuth tokens return "OAuth authentication is currently not supported" on count_tokens, so there's no way to get the headers from outside the process
• Reading from disk: The rate limit state is only held in memory, not persisted
• Parsing transcripts: Rate limit info is not written to session transcripts

None of these work — the data is only accessible inside the Claude Code process.

Environment

• Claude Code v2.1.71
• Subscription: Team (Claude Max 5x)
• Platform: Linux (Fedora 43)