[Claude cost] Usage & Spend shows one claude-swap account because the cost scan reads a single home

[Newarr]

Summary

The Usage & Spend pane prices Claude usage from one home directory only. I run two subscriptions through claude-swap, and sessions from the second slot never show up in the 30-day totals. The pane does not error, it just undercounts.

The menu bar tracks rate windows for every swap account, and #2713 extended that to the dashboard-v1 snapshot. Spend was left out of that work because claude-swap list carries no spend field. The Usage & Spend pane still has the gap.

Environment

• CodexBar 0.49.6
• claude-swap with 2 accounts, swap integration enabled
• One CLAUDE_CONFIG_DIR home per swap slot

Reproduction

1. Set up two accounts in claude-swap and enable the integration.
2. Run Claude Code sessions under both.
3. Open Settings > Usage & Spend, select 30d.

Observed behavior

• Claude shows a single row built from ~/.claude sessions. The local cost cache (claude-v6.json) holds 6,978 scanned files. Every one sits under ~/.claude/projects. The second slot's home has zero.
• cswap list reports both accounts and the menu bar shows windows for both, so the inventory is there.

Cause (source at f15f142)

SpendDashboardController builds one input per non-Codex provider from tokenSnapshotPublicationForCurrentProviderConfig(for:) (Sources/CodexBar/SpendDashboardController.swift:224-247). Codex gets per-account treatment: codexRequests walks every visible account with its own home path and cache identity (line 468). Claude has no counterpart, so every slot except the scanned home is invisible to spend.

Expected behavior

What Codex already does:

• one spend source per claude-swap slot
• each source scans that slot's CLAUDE_CONFIG_DIR
• one slot failing does not take down the others

Open question

My configured Claude directory (a non-default home) also has zero files in the cost cache. I could not tell whether the cost scan honors claudeConfigDirectory at all, or only the default ~/.claude. Worth confirming while the scan scope is open.

[clawsweeper]

Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve. Reviewed August 15, 2026, 6:27 AM ET / 10:27 UTC.

Summary
Keep open for maintainer product review: current source proves that Usage & Spend captures one ambient Claude scan, but reliably adding claude-swap homes requires a new scoped-root contract or setting that the current adapter deliberately does not provide.

Reproducibility: yes. source-reproducible: current main constructs one non-Codex Claude input and only scans the process-configured Claude root, while the adapter exposes no additional root mapping.

Maintainer decision needed

Question	Recommendation
Should CodexBar add an opt-in user-configured set of Claude project roots for Usage & Spend, or wait for claude-swap to expose a stable slot-to-root contract before offering per-slot totals?	Add opt-in scoped roots: Add explicitly configured roots with independent scans and aggregate them as local Claude spend without claiming an unproven slot identity.

Why: The current adapter intentionally accepts no local-root data and forbids direct storage reads, so source code cannot safely choose a mapping, persistence model, or attribution promise.

Next step
A maintainer must choose the scoped-root and account-attribution contract before a safe implementation can be bounded.

Review details

Best possible solution:

Approve an opt-in, privacy-preserving scoped Claude-root design—preferably aggregate-only unless an upstream stable slot-to-root contract exists—and add independent cache identities plus regression coverage for healthy and failing roots.

Do we have a high-confidence way to reproduce the issue?

Yes, source-reproducible: current main constructs one non-Codex Claude input and only scans the process-configured Claude root, while the adapter exposes no additional root mapping.

Is this the best way to solve the issue?

Unclear: per-slot scanning is not safely implementable from the current claude-swap contract; an explicit scoped-root policy or upstream versioned mapping is the maintainable path.

AGENTS.md: found and applied where relevant.

Remaining risk / open question:

• The current claude-swap schema cannot associate a slot with a Claude session directory; inventing that association could misattribute local spend or broaden filesystem access beyond the accepted adapter boundary.

Codex review notes: model internal, reasoning high; reviewed against f15f142a7787.

Label changes

Label changes:

• add P2: The report describes a bounded undercount in the existing Usage & Spend view, without a crash, credential loss, or blocked setup path.
• add impact:other: The concrete impact is inaccurate local spend accounting, which does not fit the more specific owned impact classes.
• add issue-rating: 🦞 diamond lobster: Current issue advisory state selects this label.
• add clawsweeper:source-repro: Current issue advisory state selects this label.
• add clawsweeper:no-new-fix-pr: Current issue advisory state selects this label.
• add clawsweeper:needs-maintainer-review: Current issue advisory state selects this label.
• add clawsweeper:needs-product-decision: Current issue advisory state selects this label.

Label justifications:

• P2: The report describes a bounded undercount in the existing Usage & Spend view, without a crash, credential loss, or blocked setup path.
• impact:other: The concrete impact is inaccurate local spend accounting, which does not fit the more specific owned impact classes.

Evidence reviewed

What I checked:

• Current spend capture: The dashboard adds exactly one input for each non-Codex provider from that provider's current token-snapshot publication; Claude therefore has one ambient snapshot rather than one source per claude-swap slot. (Sources/CodexBar/SpendDashboardController.swift:224, f15f142a7787)
• Configured Claude root behavior: The local scanner does honor a non-empty CLAUDE_CONFIG_DIR in the CodexBar process, scanning its projects directory; it does not enumerate additional per-slot roots. (Sources/CodexBarCore/Vendored/CostUsage/CostUsageScanner+Claude.swift:40, f15f142a7787)
• claude-swap contract boundary: The strict schema model contains slot, display email, active state, and usage windows, but no config-directory or session-root field; repository policy also says CodexBar does not read claude-swap or Claude Code storage directly. (Sources/CodexBarCore/Providers/Claude/ClaudeSwap/ClaudeSwapAccountList.swift:21, f15f142a7787)
• Related merged work: The merged dashboard account work made claude-swap windows visible but explicitly excluded spend because the list schema has no spend data; it is adjacent, not a fix for scoped local cost scanning. (Sources/CodexBarCore/Providers/Claude/ClaudeSwap/ClaudeSwapAccountList.swift:4, b1b80a9b3bc1)
• Feature history: The feature history identifies the original read-only claude-swap adapter and later dashboard account work as the relevant ownership trail. (Sources/CodexBarCore/Providers/Claude/ClaudeSwap/ClaudeSwapAccountList.swift:4, 88e8a537f266)

Likely related people:

• steipete: Authored the merged claude-swap dashboard account implementation and owns the current main-line source attribution for the spend and scanner surfaces. (role: recent area contributor; confidence: high; commits: b1b80a9b3bc1, 3a9a86c0cf00; files: Sources/CodexBar/SpendDashboardController.swift, Sources/CodexBarCore/Vendored/CostUsage/CostUsageScanner+Claude.swift)
• optimiz-r: Introduced the original read-only claude-swap account adapter whose intentionally narrow contract remains relevant to any scoped-root design. (role: introduced behavior; confidence: medium; commits: 88e8a537f266, 6c4cf3a77f57; files: Sources/CodexBarCore/Providers/Claude/ClaudeSwap/ClaudeSwapAccountList.swift)

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.