Skip to content

Releases

v1.8.0

Choose a tag to compare

@github-actions github-actions released this 18 Nov 20:58

Added

  • Why a channel failed, in the opt-in daily usage summary. Each channel's ok/fail counts now come with a reasons count of the failures by a fixed category (timeout, spawn_error, exit_nonzero, no_tty, http_4xx, http_5xx, network, unsupported, other), so a toast that never appears can be told apart from a bell with no terminal. Only the category is counted, never a message, status code, exit code, path or URL; ok and fail are unchanged and nothing is sent unless usage stats are on.
  • Skip alerts while you are looking at the agent (opt-in). New suppressWhenFocused (default false, off) in ~/.anotifier/config.json, or anotifier config focus: when it is on and the window running the agent (its terminal or editor) is the one in front, every alert (task complete, needs input, and a new session if you turned that alert on) is not sent on any channel (toast, ntfy, webhook, bell). Only an answer of "yes, in front" skips; an unknown answer (tmux, screen, zellij, ssh, Wayland, a probe that fails or times out) sends as usual. It is off by default because it needs a window probe before each alert, which adds about 1 s to each alert on Windows and WSL while it is on, and because another tab or pane of the same terminal window counts as in front. A window left in front while you are away still counts as in front, so the idle reminder is never suppressed. It can be turned on or off for one project in a projects rule. Usage-limit and context warnings and the update notice do not use it. anotifier status shows it as When focused.
  • How long the turn took. A "task complete" notification now ends with the turn length when it is known, for example app: Task complete (took 4m 12s) (also on the assistant's own words when rich content is on). Titles are unchanged. Works for Claude Code, Codex CLI and Gemini CLI, which give a hook at the start of a turn; Cursor and Antigravity CLI have no reliable start signal, so they show nothing.
  • Minimum duration filter. New events.task_complete.minDurationSeconds (default 0, off): a turn shorter than that many seconds sends no "task complete" alert. Permission prompts and every other needs-input notification are always sent, a turn of unknown length is never skipped, and the key is rejected on any event other than task_complete. Set it in anotifier config events or ~/.anotifier/config.json; anotifier status shows it as Min duration. See the README section "Turn length and the minimum duration filter".
  • A timing hook that records when a turn started: UserPromptSubmit for Claude Code and Codex CLI, BeforeAgent for Gemini CLI. It prints only {}, has a 5-second timeout, ignores every error, and keeps its state in ~/.anotifier/turns.json. Re-run npx anotifier setup after upgrading to add it (the Claude Code plugin gets it from hooks/hooks.json); anotifier uninstall removes the hook and deletes turns.json (with its lock and any leftover temp file). Until it is wired, nothing is timed and nothing is filtered.
  • Your own sound files. events.<event>.toastSound can now be the path of an audio file: ~/sounds/done.wav, an absolute path, any value with a / or a backslash, or a name ending in .wav, .mp3, .aiff, .aif, .m4a, .aac, .ogg, .oga, .opus or .flac. It plays in the background: .wav and Windows Media formats on Windows (capped at 30 seconds, toast made silent), afplay on macOS (the notification is sent without a sound name), pw-play/paplay/aplay on Linux (the only sound Linux plays), Windows playback under WSL (toast made silent). A missing file leaves the default sound on Windows (on macOS the notification is sent as it is for Default) and a player that cannot start is logged, both without the path; a file a player cannot decode stays silent on macOS and Linux. anotifier config sounds checks the file exists, anotifier test toast plays it, and anotifier doctor warns about a missing one.
  • Per-project overrides. New projects list in ~/.anotifier/config.json: each rule has an absolute or ~/ match (a folder, or a glob such as ~/work/*/app; a rule whose match is relative, such as foo, ./app or C:foo, is rejected and ignored, and a hook folder that is not absolute, such as foo, ./app, ~/x or C:foo, matches no rule) and overrides the channel (ntfy, toast, terminalBell, webhook), sound, priority, minDurationSeconds, quietHours, per-event settings or source labels for hooks run in that folder or anything inside it. Rules are applied once per hook run, before quiet hours, the minimum duration filter and routing, so all of them see the same settings; a run with no folder in its hook input, or no matching rule, behaves exactly as before. If applying the rules ever fails, the error is written to ~/.anotifier/errors.log (context config:projects) and the hook carries on with your top-level settings. Rules live only in your own config file, never in a per-repo file, so a cloned repository cannot redirect your alerts. A Telegram rule that sets format inherits chatId from the top-level webhook. sentry, updateCheck, telemetry and usageAlerts cannot be overridden. anotifier status, doctor and config still show your top-level settings only. The statusline's context and usage-limit warnings apply them too, using the folder the session is in at each statusline refresh, from Claude Code's statusline input (workspace.current_dir, then workspace.project_dir, then cwd): a matching rule's contextAlerts (enabled, threshold) changes the context warning for that session only, and its ntfy, toast, webhook and quietHours change how both warnings are delivered. The usage-limit warning is account-wide with one shared state, so its thresholds and on/off switch stay global (usageAlerts is still not allowed in a rule). A rule that turns every channel off leaves the usage warning to your other chats. See the README section "Per-project overrides".

Fixed

  • Cursor alerts now read Cursor's own hook input. Cursor sends workspace_roots, conversation_id and, on the stop hook, a status (completed, aborted or error); the hook read none of them. Now the first workspace root names the project, so a Cursor alert shows my-app · Cursor in its title and a Windows click on the toast can focus the right window. The conversation id is the session id, so the double-fire guard is per conversation instead of one global lock: two chats finishing within 1.5 seconds both alert. A run that errored now says Task failed (with the warning ntfy tag instead of the check) rather than "Task complete", still under your events.task_complete settings. The failed-run alert uses the warning ntfy tag in place of whatever events.task_complete.ntfyTags is set to. A run you aborted sends no alert at all, since you stopped it at the keyboard. Each field is used only when present and well-typed, and only for Cursor (other agents' payloads are read exactly as before); a payload without them alerts exactly as before. Windows Cursor sends workspace roots like /C:/Users/x/app; only the folder name (app) is used. Hook input that starts with a UTF-8 byte-order mark, which Cursor on Windows can send, is now parsed instead of being dropped as malformed. Nothing to re-run: setup already wires the stop hook.

Changed

  • Cursor's official app icon replaces the old cube icon (assets/icons/cursor.png), shown on Windows and Linux notifications, in the README and on the site.
  • Message text for webhook and ntfy consumers. A "task complete" message can now end with (took …) once the timing hook is wired, and Cursor alerts are titled my-app · Cursor, say Task failed when the run errored and send nothing for an aborted run. Anyone matching message text exactly should update.

What's Changed

  • docs(landing): anotifier 1.7.1, session_start is off on the webhook by default by @AminDhouib in #95
  • feat(approval): remote Allow/Deny from the phone for Claude Code Bash (experimental, not enabled by setup) by @AminDhouib in #96
  • fix(brand): official logos for Cursor, Slack, Telegram and GitHub by @AminDhouib in #90
  • feat(approval): setup warns when the ntfy account has no tier by @AminDhouib in #97
  • feat(landing): first-party analytics through anotifier.io (PostHog /_t/n4q, gtag /gx4t) by @AminDhouib in #98
  • seo(landing): Claude Code hooks security guide by @AminDhouib in #99
  • seo(landing): ccnotify guide by @AminDhouib in #100
  • fix(landing): guide overview link and icon alt use the agent's name by @AminDhouib in #101
  • feat(landing): session replay on PostHog and Sentry (10%), person profile on conversion by @AminDhouib in #102
  • fix(landing): Sentry session replay 10% → 5% by @AminDhouib in #104
  • seo(landing): deepen /vscode/ and /discord/, link them from guides by @AminDhouib in #103
  • fix(landing): Sentry session replay 5% → 10% by @AminDhouib in #105
  • seo(landing): Codex CLI guide covers Windows and turning alerts off by @AminD...
Read more

v1.7.1

Choose a tag to compare

@github-actions github-actions released this 10 Oct 17:55
1078e63

Changed

  • session_start is now off on the webhook by default, like every other channel. The default config already turned off the toast, ntfy and the terminal bell for a new session, but not the webhook, so anyone with a webhook enabled (Slack, Discord, Telegram, generic) got a "session started" post on every new agent session. Your config is deep-merged over the defaults: if you never set events.session_start.webhookEnabled, those posts stop after you upgrade; if you set it to true, nothing changes. To get them back, add "events": { "session_start": { "webhookEnabled": true } } to ~/.anotifier/config.json (the anotifier config prompts have no per-event webhook setting).

Fixed

  • On Windows, a usage or context warning lock released by another chat at the same moment is no longer logged as an EPERM error in anotifier status.

What's Changed

  • test(notify): fire the duplicate claude runs together in the dedup bell test by @AminDhouib in #89
  • docs(landing): anotifier 1.7.0, ntfy fallback server by @AminDhouib in #91
  • fix(alerts): retry the lock once when a Windows EPERM race outruns the stat by @AminDhouib in #92
  • fix(config): session_start is off on the webhook by default, like every other channel by @AminDhouib in #93
  • chore(release): 1.7.1 by @AminDhouib in #94

Full Changelog: v1.7.0...v1.7.1

v1.7.0

Choose a tag to compare

@github-actions github-actions released this 10 Oct 09:45
4ac0d45

Added

  • ntfy fallback server. New optional ntfy.fallbackServer: when the main
    ntfy.server fails (unreachable, timed out, or an error such as a 429 daily
    limit), the same push goes once to the fallback, same topic. Run your own ntfy
    server and keep https://ntfy.sh as the fallback, so a push is not lost when
    either one is down or over its limit. With a fallback set each attempt waits
    at most 3 s (instead of 5 s) so both fit in the hook budget. A fallback failure
    is logged as ntfy:fallback; anotifier status shows the fallback, and
    anotifier test ntfy warns when only the fallback delivered.

Fixed

  • ntfy timeout is a real deadline. Each ntfy send now gives up after its
    timeout in total (DNS, connect and response). The old socket idle timer
    could be kept alive by a server sending a byte at a time.

What's Changed

Full Changelog: v1.6.1...v1.7.0

v1.6.1

Choose a tag to compare

@github-actions github-actions released this 10 Oct 09:05
26e1dce

Changed

  • Windows toast failures say why. A failed Windows toast used to be logged
    as Command failed: pwsh ... toast.ps1 ..., which reads the same for a
    timeout, a non-zero exit and a missing pwsh. anotifier status now shows the
    cause (timed out after 7000 ms, exit code 3, pwsh not found), and the
    error log keeps the exit code, signal, elapsed time and stderr.

Fixed

  • Per-model "auto" auto-compact window. A modelSettings.<model>.autoCompactWindow
    of "auto" (the window Claude Code tunes for the model) now means the model's
    own window for the context warning, instead of falling through to a
    top-level autoCompactWindow number that Claude Code does not apply to that
    model.
  • No false lock errors on Windows. Two chats refreshing their statusline at
    the same moment could log usage-alert:lock EPERM: Windows reports a lock
    file that is being released as EPERM instead of "already exists". That is now
    treated as the lock being busy and skipped quietly; a real permissions problem
    is still logged.

What's Changed

  • fix(windows): log why a toast failed (timeout, exit code, missing pwsh) by @AminDhouib in #81
  • seo(landing): anotifier 1.6.0 — Claude Code auto-compact (context) warning guide by @AminDhouib in #83
  • fix(alerts): treat a Windows EPERM on a lock being released as contention, not an error by @AminDhouib in #82
  • fix(context): a per-model autoCompactWindow of "auto" means the model window by @AminDhouib in #84
  • chore(release): 1.6.1 by @AminDhouib in #85

Full Changelog: v1.6.0...v1.6.1

v1.6.0

Choose a tag to compare

@github-actions github-actions released this 09 Oct 21:26
7fbcd77

Added

  • Context warning for Claude Code. anotifier now sends one notification
    per chat when its context passes 85% of the auto-compact window, so you can
    wrap up or /compact before Claude Code compacts on its own. The window is
    resolved from CLAUDE_CODE_AUTO_COMPACT_WINDOW, then the autoCompactWindow
    setting (/autocompact; per-model and top-level, project then user
    settings, clamped to 100K-1M), then the model's own window, and never
    exceeds the model's window. Each chat warns once and never again, even after
    /compact (/clear starts a new session). It is skipped when auto-compact
    is off, and the --autocompact flag and managed settings are invisible to a
    statusline, so the warning can be late or missing there.
    It works with API-key and proxy sessions too (no subscription limits needed),
    goes to your toast, ntfy and webhook channels, and is held back by snooze and
    quiet hours like the usage warnings. Configure or turn it off with the new
    contextAlerts block (enabled, threshold); anotifier status shows it.

What's Changed

  • seo(landing): Antigravity CLI agent page + guide, wsl-notify-send guide by @AminDhouib in #75
  • Stop tests from polluting the real ~/.anotifier/errors.log by @AminDhouib in #78
  • seo(landing): anotifier 1.5.0 — Claude Code usage-limit warning guide by @AminDhouib in #77
  • feat(claude): warn once per chat when context passes 85% of the auto-compact window by @AminDhouib in #79
  • chore(release): 1.6.0 by @AminDhouib in #80

Full Changelog: v1.5.0...v1.6.0

v1.5.0

Choose a tag to compare

@github-actions github-actions released this 09 Oct 07:49
51b3afc

Added

  • Usage-limit warnings for Claude Code. anotifier now warns you before a
    Claude Code usage limit stops you: one notification when the 5-hour or the
    weekly window crosses 70%, 85% and 95% (priority default, high, urgent),
    naming the chat and when the window resets. Limits are account-wide, so each
    threshold warns once per window across all your chats. It needs a Claude.ai
    Pro/Max plan (Claude Code only reports usage to those) and goes to your
    toast, ntfy and webhook channels, held back by snooze and quiet hours.
    anotifier setup routes Claude Code's statusline through
    src/statusline.mjs, wrapping your own statusline if you have one, and
    anotifier uninstall restores it exactly. Configure or turn it off with the
    new usageAlerts block (enabled, thresholds); anotifier status shows
    it. Plugin-only installs wire the statusline by hand (see the README).
  • Where to get help. anotifier --help prints the anotifier Discord
    invite, and anotifier doctor points there whenever a check warns or fails
    (doctor --json is unchanged). The README has a Community section.

What's Changed

  • ci(tui): run F2's codex on proxyai free models instead of the OpenAI API by @AminDhouib in #71
  • seo(landing): anotifier 1.4.0 — Antigravity CLI support, notification_type routing by @AminDhouib in #72
  • ci(live-claude): run the Claude lanes and TUI F1 on the proxyai proxy by @AminDhouib in #73
  • feat(community): link the anotifier Discord from the CLI, README and every page of anotifier.io by @AminDhouib in #62
  • feat(claude): warn before the Claude Code usage limit (70/85/95% of the 5-hour and weekly windows) by @AminDhouib in #74
  • chore(release): 1.5.0 by @AminDhouib in #76

Full Changelog: v1.4.0...v1.5.0

v1.4.0

Choose a tag to compare

@github-actions github-actions released this 08 Oct 14:35
e3dfc43

Added

  • Antigravity CLI support. Google moved Google AI Pro/Ultra and free
    Code Assist users from Gemini CLI to Antigravity CLI on June 18,
    2026, and it ignores the hooks anotifier writes to ~/.gemini/settings.json.
    anotifier setup now detects it (~/.gemini/antigravity-cli) and adds an
    anotifier group with a Stop handler to ~/.gemini/config/hooks.json,
    merging beside your own groups and backing the file up first. A finished run
    sends the usual "Task complete" alert, titled <project> · Antigravity.
    anotifier uninstall removes only that group, and status reports it.
    The handler always answers {"decision":"stop"}, so it can never keep the
    agent running. Antigravity has no notification or permission event, so there
    is no "needs input" alert for it. Hook wiring and payload parsing are
    unit-tested against Google's hooks documentation; there is no live
    Antigravity CLI lane in CI yet.

Fixed

  • Claude notifications that don't need you are no longer urgent. The
    Claude Code Notification hook is registered for every notification type,
    and anotifier never read the payload's notification_type, so a completed
    login (auth_success), a finished background agent (agent_completed), an
    answered MCP form (elicitation_response) and any newer type all went out
    as the urgent "Needs your input" alarm. anotifier now routes on the type:
    permission_prompt is urgent and says "Needs your permission";
    elicitation_dialog, elicitation_url_dialog, agent_needs_input and
    quota_auto_resume_stale stay urgent "Needs your input" (or "Needs you to
    resume"); idle_prompt gets the quieter idle-reminder treatment by type
    instead of by wording; agent_completed and the other quota_auto_resume_*
    types become a default-priority notice with an info tag; auth_success,
    elicitation_response and elicitation_complete send nothing; and an
    unrecognized type is an urgent "Needs your attention" (a type added by a
    future Claude Code release may be waiting on you, so it fails loud).
    Payloads with no notification_type (older Claude Code) are routed exactly
    as before. Opt-in usage stats count the silent runs as a new skipped
    outcome.
  • Toasts on a machine with no toast backend. anotifier setup used to warn
    that notify-send (Linux), PowerShell 7 / BurntToast (Windows) or Windows
    PowerShell interop (WSL) was missing but leave the toast channel on, so every
    run then failed the toast silently while bell and ntfy delivered. Setup now
    turns the toast channel off (toast.enabled: false, recorded as
    toast.disabledBySetup: true) and prints what to install. Re-running
    anotifier setup once the backend exists turns toasts back on; toasts you
    turned off yourself are left off. anotifier status and anotifier doctor
    show "toast: off, no backend ()". Only a definite miss turns
    toasts off: a probe that times out (a slow PowerShell start, say) leaves toasts
    on, in interactive and non-interactive setup runs alike. The Linux check also
    no longer depends on the which binary being installed.

Security

  • config.json is owner-only. The config can hold webhook URLs (which are
    credentials), the ntfy topic and the Sentry DSN, but it was written with the
    default umask, leaving it world-readable (0644) on macOS and Linux. It is now
    written with mode 0600, and a newly created ~/.anotifier directory is 0700.
    Existing files are tightened the next time the config is saved.

What's Changed

  • chore(landing): site version 1.3.1 by @AminDhouib in #54
  • docs(landing): name 1.3.1 in the notification_type caveat by @AminDhouib in #55
  • fix(landing): patch vulnerable dependencies by @AminDhouib in #56
  • ci: give every workflow a least-privilege GITHUB_TOKEN by @AminDhouib in #57
  • fix: close the CodeQL JavaScript findings (config prototype pollution) by @AminDhouib in #58
  • seo(landing): answer-first titles and descriptions for three zero-click page-1 guides by @AminDhouib in #59
  • seo(landing): link every agent and channel page from the site footer, retitle the Claude Code notifications guide by @AminDhouib in #60
  • fix: route Claude notifications by notification_type, only alert urgently when Claude waits on you by @AminDhouib in #61
  • feat(landing): report-only CSP reporting to Sentry by @AminDhouib in #64
  • docs: security design for approving agent permission requests from the phone by @AminDhouib in #63
  • fix: write config.json owner-only (0600) by @AminDhouib in #65
  • perf(landing): self-host Inter and JetBrains Mono (no build-time Google Fonts fetch) by @AminDhouib in #67
  • fix: setup turns toasts off when no toast backend is installed by @AminDhouib in #66
  • test(telemetry): time the hook budget from the send, not server arrival by @AminDhouib in #69
  • feat: Antigravity CLI support by @AminDhouib in #68
  • chore(release): 1.4.0 by @AminDhouib in #70

Full Changelog: v1.3.1...v1.4.0

v1.3.1

Choose a tag to compare

@github-actions github-actions released this 06 Oct 20:19
cd2cde7

Fixed

  • Plain output off a terminal. When stdout is not a terminal (piped, CI
    logs, or an agent running npx anotifier setup through its shell tool) the
    CLI now prints plain text instead of 24-bit colour escapes, and the setup
    spinner prints only its result line instead of a run of animation frames.
    NO_COLOR forces plain text and FORCE_COLOR forces colour.
  • Banner. anotifier --help showed the pre-rename "AI Notify" ASCII art; it
    now spells anotifier.

What's Changed

  • chore(landing): site version 1.3.0 by @AminDhouib in #51
  • feat: install command at the top of every guide, answer-first snippets, plain CLI output off a terminal by @AminDhouib in #52
  • chore(release): 1.3.1 by @AminDhouib in #53

Full Changelog: v1.3.0...v1.3.1

v1.3.0

Choose a tag to compare

@github-actions github-actions released this 06 Oct 07:47
d815d42

Added

  • Documentation URL. anotifier --help and the end of anotifier setup now
    point at the full reference at https://anotifier.io/docs/ instead of the
    bare home page.
  • Support link. package.json carries the npm-standard funding field
    (npm fund surfaces it), the repo has a .github/FUNDING.yml, and the CLI
    prints a single muted line — "consider supporting it" — at the end of
    anotifier setup and anotifier status, plus a footer line in --help. The
    URL lives in exactly one place (src/support.mjs reads funding.url). It is
    never printed on the hook path, never inside a notification, and never in
    doctor --json.
  • Star ask after setup. A successful anotifier setup ends with one muted
    line pointing at the GitHub repo ("A GitHub star helps other developers find
    anotifier"). It appears only there — not in status, --help, doctor, the
    hook path, or any notification. The URL comes from package.json
    repository.url.
  • Opt-in anonymous usage stats. Off unless you say yes. The first
    anotifier setup run in an interactive terminal asks (default Yes); a setup
    that is piped, driven by an agent (/anotifier:setup), or whose input closes
    never asks and leaves stats off, and re-running setup never overrides a choice
    you already made. anotifier telemetry [on|off] shows or changes the choice,
    prints your install id and the exact counts waiting to be sent, and off
    deletes the local install id and pending counts. CLI commands send their name,
    clamped argument shape, exit code, duration and per-channel results in one
    request; hook runs are only counted locally and sent as one summary a day
    (outcomes per agent, event and channel, skip reasons, unrecognized hook event
    names clamped to 40 safe characters and capped at 10, a latency bucket, and
    which features are on). The README lists every event and property. Message
    text, project names, paths, ntfy topics and servers, webhook URLs, hostnames
    and error messages are never sent. Data goes to the project's self-hosted
    PostHog (posthog.devino.ca), keyed by a random install id, with geolocation
    disabled; the server discards client IP addresses. To delete what was sent,
    open a GitHub issue with your install id. Off by default in config.json (so
    plugin-only installs send nothing), and forced off by DO_NOT_TRACK=1,
    ANOTIFIER_TELEMETRY=0 or CI. The hook waits at most 0.8s for the server (the
    network send only; the small local counter file is updated in place and is
    not part of that wait) and the CLI at most 1s.

Fixed

  • WSL is reported as WSL. anotifier setup, status and doctor described
    WSL as plain Linux and checked for notify-send, which WSL never uses. They
    now name the Windows-interop toast path, and doctor checks for wslpath
    and a reachable Windows PowerShell instead.
  • WSL toast failures are logged. When no Windows PowerShell could be reached
    through interop, the toast failed silently. It now writes a toast:wsl entry
    (naming each executable tried) to the hook error log that status shows.
  • doctor no longer passes with only Windows PowerShell 5.1. Toasts are
    spawned through PowerShell 7 (pwsh), so 5.1 alone now fails the toast check
    with a winget install hint, and setup says so instead of reporting a
    BurntToast install failure.
  • uninstall removes the agentfocus:// protocol that click-to-focus
    registers under HKCU\Software\Classes, but only when it still points at
    anotifier's focus.vbs.
  • The generated ntfy topic uses crypto.randomInt instead of Math.random.
    The topic is what keeps a public ntfy.sh topic private.

What's Changed

  • feat: docs URL + single-source support link (CLI, README, package.json funding) by @AminDhouib in #23
  • chore: track anotifier-landing as a git submodule at ./landing by @AminDhouib in #24
  • chore: bump landing submodule to 0dd9280 (meta descriptions under 160 chars) by @AminDhouib in #25
  • chore: move the anotifier.io source into landing/ (was a private submodule) by @AminDhouib in #26
  • ci: store workflow artifacts on self-hosted MinIO (DevinoSolutions/artifact) by @AminDhouib in #21
  • ci: timeouts and cancel-in-progress for superseded runs by @AminDhouib in #27
  • Landing: Star on GitHub button with live count by @AminDhouib in #28
  • CLI: one-line GitHub star ask after setup by @AminDhouib in #29
  • fix(landing,tests): star button review follow-ups by @AminDhouib in #30
  • ci: least-privilege permissions for the five live/toast workflows by @AminDhouib in #31
  • fix(landing): SEO targeting, mobile LCP, canonical redirects + live-site tests by @AminDhouib in #33
  • ci: pin DevinoSolutions/artifact/upload to its v1 commit SHA by @AminDhouib in #32
  • fix(cli): WSL reporting + logging, pwsh-only doctor, uninstall protocol cleanup, CSPRNG topic by @AminDhouib in #35
  • test(landing): live link check — every CTA must land where it points by @AminDhouib in #36
  • feat(landing): per-page share images, copy buttons, GA events, a11y, 5 new guides by @AminDhouib in #34
  • docs(landing): version-qualify the WSL/Windows gaps #35 fixed by @AminDhouib in #37
  • test(landing): install funnel E2E — site version and command match npm by @AminDhouib in #38
  • fix(landing): send HSTS, and test security headers on every response by @AminDhouib in #39
  • docs(landing): Claude Code permission + sound guides, answer-first rework, link every guide from home by @AminDhouib in #40
  • perf(landing): star count after first paint + gzip 6 (mobile LCP 3.6→2.7 s local); per-page sitemap lastmod by @BSalaeddin in #41
  • docs(landing): Codex notification sound guide; fix built-in Codex notification and hooks claims by @AminDhouib in #42
  • test(landing): stop Chrome when the browser test can't reach it, so a slow start fails instead of hanging CI by @AminDhouib in #43
  • feat(landing): expand agent pages with exact setup, limits, and the Antigravity move by @AminDhouib in #44
  • fix(landing): stop the Codex guides competing for "ding", and link uncrawled pages by @AminDhouib in #45
  • ci(tui-proofs): report an API account error in the Codex proof as INFRA by @AminDhouib in #46
  • feat(landing): load PostHog when a project key is set at build time by @AminDhouib in #47
  • feat: opt-in anonymous usage stats for the CLI and hook by @AminDhouib in #48
  • fix(telemetry): stop replacing the state file with rename on the hook path by @AminDhouib in #50
  • chore(release): 1.3.0 by @AminDhouib in #49

New Contributors

Full Changelog: v1.2.6...v1.3.0

v1.2.6

Choose a tag to compare

@github-actions github-actions released this 10 Sep 22:31

Fixed

  • Notifications name the project again. The project name lived only in the
    body prefix (my-app: Task complete), and rich content — on by default for
    toasts and webhooks — replaces the whole body with the assistant's actual
    words. So a desktop toast read "Claude Code" plus a chat snippet, with no clue
    which project had finished; with several agents running in different repos
    that made toasts nearly useless. The project name now leads the title on
    every channel (my-app · Claude Code), where nothing rewrites it. The body
    is unchanged, byte for byte; events with no project keep the bare label.
  • ntfy titles with non-ASCII characters are RFC 2047-encoded. HTTP headers
    are bytes, so the new · separator — and any non-ASCII project directory
    name — would have arrived as mojibake or thrown before the push was sent.
    ntfy decodes the =?UTF-8?B?…?= form server-side; pure-ASCII titles are
    sent exactly as before.

What's Changed

  • fix: put the project name in the notification title so rich toasts keep it by @AminDhouib in #22

Full Changelog: v1.2.5...v1.2.6