accounts usage drops additional_rate_limits[] (per-model/feature windows not displayed)

[longfin]

Summary

accounts usage only renders the top-level 5h / weekly windows and silently drops additional_rate_limits[] from the upstream payload. As a result, per-model / per-feature limits (e.g. GPT-5.3-Codex-Spark / codex_bengalfox) are never visible in the (non---json) output, even when they are the windows actually being consumed.

Split out from #23 (which is about usage not being charged at all on the responses path); this one is purely a client-side display gap and is independently fixable.

Where

src/client/usage.ts — normalizeUsage (around L98–123) only reads rate_limit.primary_window and rate_limit.secondary_window:

return {
  plan_type: str(o.plan_type),
  allowed: bool(rl.allowed),
  limit_reached: bool(rl.limit_reached),
  primary: parseWindow(rl.primary_window),
  secondary: parseWindow(rl.secondary_window),
  credits: ...,
  raw,
};

The upstream …/wham/usage payload also contains an additional_rate_limits[] array, which is dropped here (it only survives inside raw, so --json consumers can still reach it, but accounts usage and accountsUsageCommand's human output never surface it).

Example upstream shape (observed):

"additional_rate_limits": [
  {
    "limit_name": "GPT-5.3-Codex-Spark",
    "metered_feature": "codex_bengalfox",
    "rate_limit": {
      "allowed": true,
      "limit_reached": false,
      "primary_window":   { "used_percent": 0, "limit_window_seconds": 18000,  "reset_after_seconds": 18000,  "reset_at": ... },
      "secondary_window": { "used_percent": 0, "limit_window_seconds": 604800, "reset_after_seconds": 604800, "reset_at": ... }
    }
  }
]

Impact

• For models metered under a separate bucket (e.g. the gpt-5.3-codex-spark research-preview window), accounts usage shows 0% used on the main windows and gives no indication that a different limit is filling up — misleading for monitoring / limit checks.
• Blocks verification of the serve (gpt-5.5) requests not charged against ChatGPT subscription usage — accounts usage stuck at 0% + additional_rate_limits not displayed #23 metering question: even if we route a metered codex model to confirm usage is charged, accounts usage wouldn't show the per-feature bucket moving.

Proposed fix

• Extend UsageSnapshot to carry a normalized additional?: Array<{ limit_name?: string; metered_feature?: string; primary?: RateLimitWindow; secondary?: RateLimitWindow }> parsed from additional_rate_limits[].
• Render each additional bucket in accountsUsageCommand (label by limit_name / metered_feature), and include it in usageToJson.
• Add a unit test in src/client/usage.test.ts covering a payload with additional_rate_limits.

Notes

Strictly a display/normalization gap — does not by itself explain the missing consumption in #23 (in that investigation the additional bucket was also 0). Worth fixing so usage is fully observable once the upstream accounting question is resolved.