Port upstream 0.69.0: show Claude Web limit reset credits

[Finesssee]

Summary

Claude Web source now shows the user's saved usage-limit resets ("Reset for free" in Claude Settings > Usage) as a Limit Reset Credits inventory row: N available plus the soonest expiry. It appears on the tray card, the Providers detail Usage section, and in codexbar usage text/JSON, through the existing provider-neutral ProviderInventoryItem path already used by Grok. No frontend or bridge change was needed.

• The Web usage request now asks for GET /organizations/{orgId}/usage?cedar_ember=1. If the status is not 200, 401 or 429, and it is not a Cloudflare challenge 403, the original request is retried once without the query (a surface-specific 403 may reject only the opt-in). A second failure keeps the normal error handling.
• cedar_ember is decoded by a new provider-owned module, rust/src/providers/claude/reset_credits.rs, with #[derive(Deserialize)] structs. Only eligible: true yields an inventory. A grant counts when it is not paused, resets_left > 0, started, and not expired; usable_now is not consulted. A malformed grant (bad types, missing paused, unreadable bound, resets_left outside 0..=resets_total) is dropped without hiding the rest. More than 50 available resets or more than 200 grant records shows nothing. An empty inventory shows nothing. The usage windows are never affected.
• Grant ids (redemption handles) are never deserialized. The inventory is transient (#[serde(skip)]), so nothing enters persisted usage JSON.
• Web source only. OAuth, CLI and Admin sources are untouched, and Windows has no Web-extras merge into OAuth results.
• ClaudeWebApiFetcher gains a private base_url field (default unchanged) so the wire tests can target a mock server.

Upstream reference

• Release: v0.69.0, "Claude: show saved usage-limit resets and their expiry from the Web source in the menu and codexbar usage details" (Show Claude limit resets from the Web source steipete/CodexBar#4048).
• Tag-pinned sources (v0.69.0): Providers/Claude/ClaudeRateLimitResetCredits.swift, Providers/Claude/ClaudeWeb/ClaudeWebAPIFetcher.swift (requestUsage, usageStatusesKeptWithResetOptIn, parseResetCredits), Tests/CodexBarTests/ClaudeRateLimitResetCreditsTests.swift, docs/claude.md. Wire shape and fixtures are taken from the Show Claude limit resets from the Web source steipete/CodexBar#4048 diff and its tests.

Ported / Deferred

Ported: opt-in query with single plain retry, cedar_ember parsing rules and caps, count and soonest-expiry display, CLI text/JSON row, Web-only scope, live-only (non-persisted) inventory.

Deferred / intentionally different:

• Upstream renders one expiry per reset in the menu card (2d · 6d). Win-CodexBar's shared inventory row shows the count with the soonest expiry only, which is what the Grok row already does. Per-reset expiry rows would need a new shared inventory shape; not done here.
• Upstream re-filters expired resets at render time (availableExpirations(at:)). Here the inventory is computed at fetch time; a reset that expires before the next refresh disappears on that refresh.
• Upstream's uncached HTTP client (skips the URL cache so the raw body is not kept on disk) has no equivalent need: the reqwest client used here has no on-disk cache.
• The .github/pr-proof log, CHANGELOG entry and Swift gatekeeper-test edits are upstream-only.
• Bounds are parsed with RFC 3339; a zone-less timestamp is treated as unreadable (grant dropped), matching "supplied but unreadable bound is malformed".

Validation

Toolchain cargo +1.98.0, E-core pinned (wrapper recreated locally because the shared port-audit scratch directory was wiped mid-run; same CPUs 16-31 / BelowNormal via start /affinity FFFF0000 /belownormal).

• cargo +1.98.0 fmt --all: clean
• cargo +1.98.0 clippy --workspace --all-targets -- -D warnings: pass
• cargo +1.98.0 test -p codexbar providers::claude:: -- --test-threads=4: 178 passed, 0 failed, 1 ignored (pre-existing; needs native Claude Code). This includes 8 new reset_credits unit tests and 6 new mockito wire tests (reset_opt_in_tests: opt-in success no retry, 400/403/404/422/500/503 retry once, retry failure keeps error handling, 401/429 not retried, Cloudflare challenge not retried, ordinary 403 retried then AuthRequired).
• Shared code (core, settings, spend) untouched, so the full cargo test -p codexbar and the Tauri crate/frontend suites were not re-run.

File sizes: web_api.rs was already over 1000 lines (1335) and grows to 1390; all new logic lives in the new reset_credits.rs (134 lines) and sibling test files.

Affected areas

• Claude provider (Web source)
• Tray card / Providers detail row (existing inventory rendering, no frontend change)
• CLI codexbar usage (existing inventory rendering)
• Settings / float bar / installer

UI proof

Pending: coordinator will capture CUA proof on a fresh build. A live Claude Web session with saved resets is needed; otherwise seed via the unit-level fixtures.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Finesssee]

Thermo-nuclear review

Reviewed by Codex gpt-6-luna (xhigh); verified and validated by Claude

Head reviewed: 459d367

Findings:

• P3, rust/src/providers/claude/web_api.rs:582: the plain retry goes through a pass-through helper (get_plain_usage) and repeats the same request across match arms. Style-only; no behavior defect found.

Nothing else found: provider logic stays siloed, no new dependencies, no secrets logged, no bridge or locale drift.

Follow-up (same head, no commit pushed):

• Codex's proposed consolidation of the P3 did not compile (E0382: opted_in is moved by bytes() inside the match), so I reverted it. The existing helper is correct and behavior-neutral, so the item is left as is.
• Validated at head: cargo +1.98.0 clippy --manifest-path rust/Cargo.toml --all-targets -- -D warnings passes. No frontend or UI changes.

[Finesssee]

CUA proof

Build commit: 459d3677fa23297776cdd5ba93e191340011689a (PR head) plus uncommitted, reverted proof-only patches (text in proof\689\proof-only.diff):

• Cargo.toml/Cargo.lock: [patch.crates-io] dirs shim redirecting home/config dirs under CODEXBAR_PROOF_HOME (isolated profile).
• rust/src/providers/claude/web_api.rs: fetcher base_url read from env CODEXBAR_PROOF_CLAUDE_BASE_URL (default unchanged; no TLS/URL validation changed).

Commands: pnpm --dir apps/desktop-tauri run tauri:build:debug, then proof\689\launch.sh trayPanel (real Claude Web fetch path against a local python mock on 127.0.0.1:18689, dummy CLAUDE_AI_SESSION_KEY, settings claude_usage_source=web, theme auto) and MOCK_MODE=reject proof\689\launch.sh trayPanel. Driven and captured with cua-driver (background get_window_state only; window kept on the second monitor).

Fixture grants: 2 resets (ends +5d), 1 reset (ends +2d1h), one paused grant, one expired grant.

#	Assertion	Result
1	No real email/account from the user's machine visible (only dummy proof-user@example.invalid)	PASS
2	Tray Claude card shows "Limit Reset Credits: 3 available" with "Next expires in 2d 0h" (paused and expired grants excluded)	PASS
3	Mock log shows GET .../usage?cedar_ember=1 from the real fetch path; grant id never visible	PASS
4	Session (42%) and Weekly (18%) windows still render	PASS
5	Opt-in rejected with 403: log shows opt-in then a single plain GET .../usage retry; windows render, no credit row, no error	PASS
6	Theme stays dark under auto	PASS

Screenshots (local, not committed): C:\Users\FSOS\AppData\Local\Win-CodexBar\port-audit\proof\689\shots\tray-optin.png, ...\shots\tray-reject.png. Mock logs: proof\689\mock-optin.log, proof\689\mock-reject.log.