Add Command Code provider#138
Conversation
✅ AI Code Review CompletedReview finished. Check the PR for inline comments. 📋 View Logs | 🤖 Model: |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3bb1ba4b65
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| SubscriptionPreset(name: "Go", cost: 10), | ||
| SubscriptionPreset(name: "Pro", cost: 30), | ||
| SubscriptionPreset(name: "Max", cost: 150), | ||
| SubscriptionPreset(name: "Ultra", cost: 300) |
There was a problem hiding this comment.
These preset amounts are the included credit buckets, not the monthly subscription fees, so selecting a Command Code preset will overstate monthly spend in subscription totals (for example, Go is treated as $10/m instead of the plan price). This directly skews the Quota Status: $.../m and aggregate status-bar cost for users who rely on presets instead of manual custom pricing.
Useful? React with 👍 / 👎.
![op-gg-ai-devops[bot]](https://avatars.githubusercontent.com/in/2680015?s=60&v=4){kind=small}
There was a problem hiding this comment.
honestly, this is a clean integration. you hit literally every cross-cutting file the team's playbook calls out — ProviderIdentifier (rawValue/displayName/shortName/iconName), ProviderManager + CLIProviderManager, SubscriptionSettings, ProviderMenuBuilder, StatusBarController.usagePercentCandidates + quota order + icon switch, both status bar icon views, and the xcodeproj. that's the kind of "i read the codebase before opening the PR" energy i wish more contributors had.
the cookie discovery story is also nice — env var override → OpenCommand secrets → browser cookie extraction (with Comet added), with a sensible OpenCommand local proxy short-circuit in front of the direct API. fallback ordering is well thought out.
tests are pulling their weight too: cookie parsing variants (Secure / commandcode_prod_ / bare token / rejected), CodexBar-compatible payload, OpenCommand proxy payload, and cent-precision conversion. solid.
a few things worth tightening before merge — none are blockers, so this is APPROVE with notes:
- Duplicate
import Foundationat the top ofCommandCodeProvider.swift. just a copy-paste artifact. unknownPlanaborts the whole fetch when Command Code ships a new plan id. right now if subscription isactiveand the planID isn't in your hardcodedCommandCodePlanCatalog, the provider throws and shows nothing. the moment Command Code adds a new tier (e.g.team-pro,individual-business), users on that plan get a black-hole provider until you ship an app update. should degrade gracefully — fall back to "unknown plan, remaining credits = $X" instead of failing the whole result.- Dead fields in
CommandCodeUsageSnapshot:premiumMonthlyCreditsandopensourceMonthlyCreditsare parsed and stored but never read anywhere downstream.AGENTS.mdis explicit about not keeping dead code "even for minimal patch purposes." snapshotFromOpenCommandUsagethrows when bothmonthly_limitandmonthly_spend + remainingare 0 — that's a legitimate state for a brand-new account with no spend yet. consider treating it as a 0%-used result with no plan instead of a hard parse error.
also a forward-looking nudge: this is the first provider using a hardcoded SF Symbol ("command") as its icon rather than an asset. fine as a placeholder, but if Command Code stays on the roster long-term, a proper asset would match the rest of the providers visually.
nice work overall — ship it after fixing the import dupe + the unknownPlan brittleness and we're golden.
Reviewed with anthropic/claude-opus-4-7|high
| @@ -0,0 +1,491 @@ | |||
| import Foundation | |||
| import Foundation | |||
There was a problem hiding this comment.
duplicate import Foundation — almost certainly a copy-paste slip from auto-completion. just delete this line.
import Foundation
import os.log| if let planID = subscription.planID, | ||
| subscription.status?.lowercased() == "active", | ||
| plan == nil { | ||
| throw CommandCodeProviderError.unknownPlan(planID) |
There was a problem hiding this comment.
this is the biggest concern in the PR — when Command Code adds a new plan id (think team-pro, individual-business, anything you haven't hardcoded into CommandCodePlanCatalog), and a user's subscription status == "active", this throw bubbles all the way up via ProviderError.providerError(...) and the entire Command Code row goes dark until you ship a new app build. that's a guaranteed support ticket the day Command Code expands their plan lineup.
graceful degradation pattern (similar to how ClaudeProvider handles unknown plans):
if let planID = subscription.planID,
subscription.status?.lowercased() == "active",
plan == nil {
commandCodeLogger.warning("Unknown Command Code plan id: \(planID, privacy: .public)")
// Continue with plan = nil. The result still shows remaining credits and
// the plan id string in `planType`, just without an entitlement total.
}then in makeResult, when monthlyCreditsTotal is nil, the existing fallback totalUSD = max(snapshot.monthlyCreditsRemaining, 0) already produces a usable (if 0%-used) result. user sees their balance, you get a warning log to fix later, no outage.
bonus: pass the raw planID into planType (or details.planType) so the menu shows the unknown plan name instead of "active" from subscriptionStatus.
| guard monthlyLimit > 0 else { | ||
| throw CommandCodeProviderError.parseFailed("monthly_limit is missing") |
There was a problem hiding this comment.
this guard throws parseFailed("monthly_limit is missing") when both monthly_limit and monthly_spend + remaining are 0 — but that's a totally legitimate state for a brand-new OpenCommand account that hasn't spent anything yet and is on a 0-credit free tier. throwing here turns "new user, no usage" into "provider broken" in the menu.
suggestion: treat that as a valid zero-usage result instead of a parse failure.
if monthlyLimit <= 0 {
// No spend, no remaining, no limit configured yet — degrade gracefully.
let plan = CommandCodePlan(id: "opencommand", displayName: "OpenCommand", monthlyCreditsUSD: max(remaining, 0))
return CommandCodeUsageSnapshot(
monthlyCreditsRemaining: max(remaining, 0),
purchasedCredits: 0,
premiumMonthlyCredits: 0,
opensourceMonthlyCredits: max(remaining, 0),
plan: plan,
billingPeriodEnd: resetDate,
subscriptionStatus: "OpenCommand",
authSource: authSource
)
}parseFailed should be reserved for "API returned something we genuinely don't understand," not "user has $0 of usage."
| let premiumMonthlyCredits: Double | ||
| let opensourceMonthlyCredits: Double |
There was a problem hiding this comment.
premiumMonthlyCredits and opensourceMonthlyCredits are stored in the snapshot and parsed out of the credits payload (and the test even asserts the API shape includes them), but nothing downstream reads them — not makeResult, not the menu builder, not the status bar. AGENTS.md is explicit:
Don't keep dead code even if it's minimal patch purpose.
two options:
- drop both fields from
CommandCodeUsageSnapshot+CreditsPayload+ the JSON parser, and trim the test fixture. lightest patch. - wire them up if the plan is to display "premium vs opensource credits remaining" later — but right now there's no consumer, so option 1 is honest.
1 participant
Overview
This PR adds Command Code as a first-class quota provider in OpenCode Bar.
Command Code usage can now be loaded through OpenCommand when its local proxy is running, with a direct Command Code billing API fallback that matches CodexBar's implementation.
Changes
/healthzand/v1/account/usage.commandcode.aisession cookies.Validation
ENABLE_DEBUG_DYLIB=NO.