Skip to content

CHANGELOG.md

Latest commit

 

History

History
157 lines (131 loc) · 7.92 KB

CHANGELOG.md

File metadata and controls

157 lines (131 loc) · 7.92 KB
Raw

Changelog

All notable changes to this project should be documented here.

This project uses version tags such as v0.2.0. Release binaries are published through GitHub Actions and GoReleaser.

v0.9.0

  • New limitping redeem spends a banked Codex rate-limit reset credit (with --dry-run to see which one first). Setting auto_redeem = true under [codex] lets watch and continue spend one on their own once it is close to lapsing — within 24h with usage worth reclaiming, or in its final hour. Redeeming is irreversible, so it stays off by default; the backend refuses with "nothing to reset" when no window is eligible, so a credit is never burned for nothing.
  • Reset and expiry times in status / bg status now carry their UTC offset (e.g. resets in 3h14m (Sun 00:10 UTC+8)), so a time read from a log or a screenshot taken in another zone is unambiguous. The offset is used rather than the zone abbreviation because abbreviations collide (CST is both China and US Central).
  • Claude usage 429s are now disambiguated with Anthropic's free token-counting endpoint: an explicit subscription-auth denial gets an actionable subscription-access error, while real or inconclusive rate limits retain the original 429. Claude pings also no longer report success when the interactive CLI displays the same access-disabled error.

v0.8.0

  • Reset credits in status / bg status now show the remaining lifetime until each unredeemed credit expires (e.g. expires Jul 27 07:50 (in 11d20h)), so a banked reset about to lapse is visible at a glance.
  • Localized the text output of status, bg status, and ping for Chinese locales (window lines, reset credits, weekday names, error lines), and fixed locale detection to honor POSIX precedence so LC_ALL=en_US overrides LANG=zh_CN. English output is unchanged.
  • Adapted Codex to the weekly-only limit regime introduced on 2026-07-12 (the 5h limit is temporarily removed): usage windows are now classified by their length instead of their position in the response, a missing window renders as "not currently enforced" (its key is omitted from status --json), and watch pings at the weekly reset instead of every 5h while no 5h window is enforced. The reset-credit count embedded in the usage response is used as a fallback when the detail endpoint is unavailable.

v0.7.0

  • Added limitping schedule [provider], a foreground scheduler that runs pings on fixed intervals (--every 5h) or at one or more daily local times (--at 05:00 --at 13:00, or comma-separated values). It reuses the existing ping path and supports --dry-run.
  • Added best-effort Codex reset credit reads from the Codex backend reset-credit endpoint. When available, status and status --json now report available reset credits without failing the usage read if that private endpoint is unavailable.
  • Added usage_display = "used" | "remaining" for text status / bg status, and added remaining_percent to JSON window output so scripts can consume both views.

v0.6.0

  • Added limitping continue <provider>, an interactive proxy that launches the provider's real CLI through a PTY (your terminal passes straight through) and auto-injects a continue message the moment the 5h limit recovers, so a parked long task resumes itself instead of waiting for you. Flags after the provider pass through verbatim (e.g. continue codex --yolo). It only fires on a genuine recovery edge and respects the weekly limit (weekly_threshold, credits included), writing a diagnostic timeline to continue.log. The resume message is the new per-provider continue_prompt config key (default "continue"). Unix only for now (needs a PTY).
  • Shared the "weekly window exhausted" rule between the scheduler and the continue proxy via usage.Usage.WeeklyExhausted, so both honor weekly_threshold and usable credits identically.

v0.5.0

  • Added status --json, which emits each provider's 5h/weekly usage, plan, credits, and reset timing as a JSON array for scripts and dashboards. Progress output is suppressed so stdout stays a single valid document, a provider that fails to read becomes {"provider": ..., "error": ...}, and -v embeds the raw response under raw.

v0.4.2

  • Fixed Codex pings to use the interactive, TTY-backed Codex CLI instead of headless codex exec, so the ping anchors the subscription-backed 5h window.
  • Updated Codex trigger docs and examples to match the interactive command path and clarify that per-ping token/cost output is not available in this mode.

v0.4.1

  • Fixed Claude/Codex usage reads to match the official client request shape more closely, including provider-specific headers and Codex chatgpt_base_url handling.
  • Added status-aware handling for usage endpoint 429s so watch pauses reads instead of repeatedly retrying a rate-limited endpoint.
  • Fixed usage reads on networks where Go's HTTP/2 client path returns EOF or malformed responses by using a dedicated HTTP/1.1 usage client.

v0.4.0

  • watch now draws a live status line on an interactive terminal: a spinner plus each provider's current state and a live countdown to its next ping, redrawn beneath the scrolling log. It auto-disables when output isn't a TTY (e.g. bg's log file or a pipe), so logs stay free of control sequences.
  • Added limitping background (alias bg) to run watch as a detached background process, freeing the terminal: bg start [provider] [--dry-run] launches it, bg status (or bare bg) reports pid/uptime/log path plus the resolved list of watched providers and each one's current 5h/weekly usage, bg stop ends it, and bg logs [-f] [-n N] shows its output. Only one background watcher runs at a time; state and logs live under the config dir (bg.json / bg.log).
  • Fixed the Claude trigger: the interactive session now actually submits the prompt (and waits for the turn to run) instead of exiting before any message was sent, so the 5h window reliably starts.
  • Added hook-based active-session detection: limitping hooks install / uninstall registers limitping's hooks in ~/.claude/settings.json and ~/.codex/hooks.json so watch can tell when a session is genuinely mid-turn (between a prompt and its Stop) rather than just having a live process. The install script sets the hooks up automatically. Without hooks, limitping skips the active-session check and pings as soon as the window resets — there is no process-list fallback (it produced false positives from unrelated Claude/Codex agent processes).
  • Removed the experimental GLM (Zhipu / Z.ai) provider. limitping now targets Claude Code and Codex only; the [glm] config block and glm provider argument are gone.

v0.3.0

  • Added short command aliases such as ping / p, status / s, watch / w, version / v, upgrade / up, and uninstall / rm.
  • Updated help output to show command aliases inline and clarify accepted ping / watch provider arguments.
  • Added Chinese CLI help text when the system locale is Chinese.
  • Documented upgrade, uninstall, and command aliases in the English and Chinese READMEs.

v0.2.1

  • watch now defers automatic pings while a Claude/Codex CLI task is already running, letting that task naturally start the next 5h window.

v0.2.0

  • Switched Claude triggering to the interactive Claude Code CLI so subscription window pings keep working after headless print mode moves to Agent SDK/API billing.
  • Added retry handling for transient usage endpoint failures and removed duplicate status error output.
  • Added limitping upgrade / limitping update to update the installed binary from the latest GitHub release.
  • Added limitping uninstall, which removes the binary and config/cache by default, with --keep-config to preserve config/cache.
  • Added open-source governance, security, privacy, and contribution guidance.

v0.1.0

  • Initial public release target.