[shuv1337]

Summary

• Add read-only Claude limit-reset enrichment per account, alongside ordinary usage.
• Show the total plus New-model/weekly (and session/full/Other) groups with separate availability and expiry dates in TUI details, static cards, compact output, and typed JSON extras.
• Preserve spend/usage on optional failures, keep credential ownership unchanged, and add no runtime dependencies.

Verified data source and safety

GET https://api.anthropic.com/api/oauth/usage?cedar_ember=1&skip_spend=1, discovered in the installed Claude Code 2.1.319 bundle. The usage-client User-Agent is required; without it, the response is ineligible with reason surface.

Fields: cedar_ember.eligible, grants[].resets_left, ends_at, starts_at, type, gives_back_weekly, clears, and paused. gives_back_weekly: true takes precedence over type: full and uses the neutral New-model label; Fable is used only when the API explicitly names it.

The client uses GET for status and a separate POST route for redemption. No redemption route was called or implemented. External credentials are never refreshed or written. All probes were GET; a guarded live CLI check additionally rejected any non-GET fetch. Details and fixture provenance: docs/claude-limit-resets.md.

Live worktree output

Claude · work
  Weekly: 92% left, resets 09:59 Sun
  Fable weekly: 98% left, resets 09:59 Sun
  Limit resets: 4
  Fable resets: 3 · expire Mar 8, 2027, 23:59 PST
  Weekly reset: 1 · expires Dec 15, 2026, 20:59 PST

Validation

• bun test: 785 passed, 0 failed (20 more tests for the review fixes).
• bun run preflight: passed on the clean committed tree.
• bun run release:pack: passed, lib/ and web/ included.
• git diff --check: passed.
• Worktree CLI (node ./shuvquota.js claude work --compact --no-color): real grants and expiry dates confirmed.
• Tests cover optional HTTP/JSON/network failures, GET-only reads, per-account isolation, external-file immutability, grouped counts/expiries, unsupported plans, privacy, and JSON/compact/static/watch integration.

No merge requested; main checkout left untouched.

Review fixes

Each requested fix is a separate commit, with a full bun test pass before committing:

1. 5246599 — Cache successful reset reads for 15 minutes per SHA-256 token hash; honor reset 429 cooldowns/Retry-After and primary usage cooldowns.
2. 3f355d1 — Five-second optional reset HTTP timeout, preserving ordinary usage on abort.
3. 27c5d61 — Centralized User-Agent; lazy cached shell-free claude --version detection with 1.5-second timeout, fallback/minimum 2.1.319. Documented stale-UA blank sections.
4. 335e1a1 — 24-hour reset clocks using hourCycle: "h23", including midnight.
5. 18f58e4 — Count future grants; preserve safe startsAt and show “from ”.
6. 239f5ca — Count unclassifiable grants under Other; neutral New-model labels unless the API explicitly names Fable. Includes an untyped future grant test.
7. 0d09391 — Invalid display zones retry with local time.
8. a3ec130 — README says “Pacific time (the renderer's display time zone)”.
9. c3993d0 — Footer/help/docs use “Up next” for upcoming quota resets.

Final local validation: 785 passed, 0 failed; clean-tree bun run preflight, bun run release:pack, and git diff --check all passed. Zero runtime dependencies retained.

Final real-account worktree check: node --import <GET-only-guard> ./shuvquota.js claude work --compact --no-color (the worktree's sq entry point). Exactly two GETs to Claude usage; the guard rejects all other methods/endpoints. The live output above is current; Fable is explicitly named by this account's API response. No credentials refreshed and no reset applied or redeemed.

CI verified green at c3993d0: test and both Socket Security checks passed. PR remains open; do not merge.