Dev Notifier 2.0.0-beta.1

Menu-bar / tray notifier for Jira, GitHub & PagerDuty (macOS + Windows).

Download & Install

Platform	File	How to install
macOS (Apple Silicon)	DevNotifier-2.0.0-beta.1.dmg	Open, drag into Applications, then see the note below
Windows	DevNotifier-2.0.0-beta.1-setup.exe	Run the installer (per-user; SmartScreen: More info -> Run anyway)
Windows (portable)	DevNotifier-2.0.0-beta.1-portable.exe	Just run it

Verify file integrity via the attached SHA256SUMS.txt (the in-app updater does this automatically).

macOS note (first launch and after every update)

This is a free, open-source app without an Apple Developer certificate,
so it is not notarized and macOS blocks it with "is damaged and can't be
opened". Clear the quarantine flag once in Terminal:

xattr -dr com.apple.quarantine /Applications/DevNotifier.app

or use System Settings -> Privacy & Security -> Open Anyway after the
first attempt. (On macOS 14 and earlier, right-click -> Open also works.)

---

What's changed

Added

• PagerDuty on-call shift reminders: a heads-up before your shift starts (default 1 day and 1 hour before, pagerduty.oncall_remind_before_minutes), when it starts, and when it ends (silent), sourced from /oncalls. Turn off with pagerduty.oncall_reminders: false
• a PagerDuty ▸ menu (when enabled) showing whether you are on-call right now and until when (or when your next shift starts), and the open incidents assigned to you — click one to open it. The Status line also shows · on-call
• PagerDuty now notifies when an incident is escalated or reassigned to you, when someone requests you as a responder, when a note mentions you, and on notes, priority/urgency changes, snoozes, ack timeouts and exhausted escalation paths on incidents assigned to you or your teams; each notification says who did it (e.g. [assigned to you] escalated to you by Bob — Disk full)
• low-urgency PagerDuty incidents are tagged · low urgency and notify silently; set pagerduty.low_urgency_sound: true to keep the sound. pagerduty.notify_team_incidents: false limits notifications to incidents assigned to you (plus escalations / responder requests aimed at you)
• Jira changes made by apps/automation rules (e.g. "Automation for Jira" moving a ticket to Code Review right after you opened a PR) no longer notify, unless the rule assigns the issue to you. New jira.suppress_automation key (default true) turns this off
• github.ci_notify (default ["fail"]) picks which CI roll-up states on your own open PRs pop up. pending — which every push of yours produces — is now remembered silently instead of notifying; add "pending" to hear about it again, or set [] to silence CI roll-ups entirely
• failing sources are no longer silent. The Status ▸ submenu shows when the last check ran and, per source, the live error (Jira: ⚠ HTTP 401 unauthorized — check the API token / username, timed out, TLS certificate verification failed (corporate proxy?), …) instead of a misleading ✓ Ready; a ⚠ badge appears next to the tray icon while a source or the config has problems; a "check problems" notification tells you which source failed and why (once per distinct failure on the timer, always on Check now); and Check now no longer claims "You're all caught up" when a source failed. Status ▸ also gained View log
• Pause notifications ▸ (1 hour / 4 hours / until resumed) mutes pop-ups without quitting; items still land in Recent and a ⏸ badge shows while paused. The pause survives restarts
• burst protection: when one check finds more than 5 new items (first launch, back from a weekend offline) they are folded into one summary notification ("12 new updates — Jira 7 · GitHub 5 — see Recent") instead of a machine-gun of sounded pop-ups; every item is still listed under Recent
• config problems are visible: unknown keys (typos such as supress_self) and values of the wrong type are listed under Status ▸ as Config: ⚠ … and logged once, instead of being silently ignored; an unparsable file is shown there too
• the "setup needed" / "Nothing to check" / dependency-check notifications carry an Open button that opens config.json
• macOS: if the app was moved since Start at login was enabled, the LaunchAgent is re-pointed at the current location on the next launch; enabling it while running from the mounted DMG is refused with an explanation (that path vanishes when the image is ejected); failures to toggle it are reported instead of silently leaving the checkbox unchanged
• poll.interval_seconds edits take effect on the next config reload (no relaunch)
• the PagerDuty ▸ menu lists every escalation level you are currently on with level-specific wording — On-call now (level 1), Backup on-call (level 2), Fallback on-call (level N) — plus your next shift; each row opens its schedule / escalation policy in PagerDuty so the menu can be checked against the service page. A schedule-less entry is labelled "direct policy target, no end". pagerduty.oncall_max_level (default 1) decides which levels count as "on-call" for the header, the Status line and the shift reminders
• pre-release builds: tagging vX.Y.Z-<suffix> (e.g. v2.0.0-beta) publishes a GitHub pre-release, which is excluded from /releases/latest so existing installs are not offered it by the in-app updater; the CHANGELOG [Unreleased] section is left in place for the final release. The updater now ranks a pre-release below its final version, so 2.0.0-beta users are offered 2.0.0 (and 2.0.0 users are never offered the beta)

Changed

• the poll cursor is only advanced when every enabled source fetched successfully. Previously a failed fetch (401, timeout, VPN not up yet after wake) looked like "no news" and the window moved on, so the events from that interval were lost for good; they are now fetched on the next successful check (fingerprints prevent duplicates from the sources that did succeed)
• Jira, GitHub and PagerDuty are fetched concurrently and in isolation: a slow source no longer delays the others, and an unexpected exception in one source is recorded as that source's error instead of aborting the whole poll and discarding the results already fetched
• Check dependencies (and Status ▸ Re-check now) re-reads config.json first, so the documented "fill in the file, save, click Check dependencies" flow works immediately instead of after the next 5-minute poll
• the unconfigured-state reminder ("Nothing to check") is shown once per distinct problem on the timer instead of every 5 minutes, and is always shown on a manual Check now (which used to end in silence)
• the in-app updater now requires a matching SHA-256 from the release's SHA256SUMS.txt; a download whose checksum file cannot be fetched, or that is not listed, is discarded with an explanation (previously it was opened unverified with only a log line). The macOS menu item is Download update… and the follow-up says what to actually do (quit, drag-replace, relaunch, and the xattr line for Gatekeeper) instead of "Installer opened — follow the installer"
• Status ▸ distinguishes GitHub: ⚠ Needs gh CLI from ⚠ Needs login (gh auth login)
• config.json is written atomically and created user-readable only (0600); menu actions (theme, skip version) patch just the key they change via config.save_config_patch and refuse to write while the file is unparsable
• the gh login lookup is skipped entirely when GitHub is disabled (it logged a WARN on every poll for non-GitHub users)
• docs: install steps lead with the xattr / "Open Anyway" path (right-click → Open no longer works on macOS 15), note the Apple-Silicon-only DMG, the "Background Items Added" notice and first-launch scan delay; new Updating section and expanded troubleshooting (permission denied, Focus allow-list, body-vs-Open click, stale login item, two icons, config typos, HTTP 401/timeout/TLS); Windows uninstall covers the installer; README gained a Known limitations section; Python requirement corrected to 3.9+; scripts/doctor.sh checks quarantine, architecture, JSON validity and a stale LaunchAgent path; the DMG "READ ME FIRST" and release-notes template were refreshed
• GitHub ci_activity threads ("workflow run failed for … branch") are suppressed under github.suppress_self: by GitHub's definition they are workflow runs you triggered, and CI on your open PRs is already covered by the roll-up items (so a failure no longer pops up twice)
• the Jira identity lookup (/rest/api/3/myself) is cached for 6 hours and, if a refresh fails, the last known id is reused; the GitHub login auto-detection is cached too
• PagerDuty incident notifications are now driven by the incident timeline (/log_entries) instead of polling incident status: every event has a stable id, so each acknowledge / resolve / escalate / reassign notifies exactly once, transitions that don't change the status (reassign, escalate) are no longer missed, and resolves of incidents assigned to you are detected even when you have no teams or the incident is old. A new incident's trigger + initial assignment is reported as a single notification
• pagerduty.suppress_self now applies to incidents assigned to you as well (previously only to team incidents), so acknowledging your own incident no longer pops up a notification
• the PagerDuty identity lookup (/users/me) is cached for 6 hours instead of running on every poll when user_id/team_ids are left blank, and incident/timeline/on-call queries follow pagination instead of stopping at 50 results
• the "All good" dependency-check and "setup needed" messages now mention PagerDuty (previously only Jira/GitHub); the TUTORIAL documents suppress_self and the new PagerDuty options
• Windows releases now ship a real installer, DevNotifier-<version>-setup.exe (Inno Setup; per-user, no admin rights): Start Menu entry, uninstaller, an optional "start when I sign in" task, and in-place upgrades that close the running app and relaunch it. The bare one-file exe is still published as DevNotifier-<version>-portable.exe. The in-app updater prefers the installer asset
• Windows builds carry the release version (bundled APP_VERSION stamp plus a VERSIONINFO resource visible in Properties ▸ Details) and ship a proper multi-size .ico
• gh is also looked up in its usual Windows install locations (Program Files\GitHub CLI, scoop shims), and the "gh CLI is not installed" hint suggests winget install --id GitHub.cli on Windows instead of brew install gh

Fixed

• your own actions could still notify you, and others' actions on your PRs could be hidden:
  • GitHub decided "who acted last" from subject.latest_comment_url, which GitHub only updates for comments. After a push, review, review request, merge or close it is stale or points at the PR itself — whose user is the PR author. So a reviewer approving your PR looked like your own activity (hidden), while your own push (or any lookup failure) fell through and popped up as "your PR updated". The latest actor is now read from the PR/issue timeline (resolving pushes to the commit's GitHub login), the login comparison is case-insensitive, and lookups are cached per thread update
  • pushes whose git email GitHub cannot link to an account (typically a mistyped git config user.email) are attributed via your profile's public email or the new github.emails list, and — on a PR you authored — assumed to be yours (a one-time hint is logged)
  • when the actor lookup failed (network hiccup) the thread was notified anyway and marked seen. It is now skipped and re-checked on the next poll instead
  • one failed /rest/api/3/myself call (it ran on every poll) switched Jira self-suppression off for that whole poll, letting your own comments/status changes through — 52 such polls were found in one log. See the identity cache above
  • Jira's legacy issue-level mode (event_mode: false) ignored suppress_self entirely; issues whose only in-window activity is yours are now skipped
• Windows builds always reported version 1.3.0 because the release version was never stamped into the exe, so every launch announced an "update available", and "Download & Install" merely opened the downloaded one-file exe — a second copy of the app — while claiming an installer had been opened. The app now reads the version stamped at build time, so it recognises when it is up to date, and updates run the real installer
• "Start at login" on Windows pinned the registry Run entry to whatever path the exe was launched from (e.g. Downloads\DevNotifier-1.5.8.exe), silently breaking once the file was moved or deleted. It now points at the installed DevNotifier.exe when the app is installed, and the installer repoints an existing entry
• gh output is decoded as UTF-8 (it was decoded with the Windows locale codec, so a non-ASCII PR title or user name raised UnicodeDecodeError and dropped the whole GitHub poll), and gh is run without flashing a console window on Windows
• GitHub threads notified once per thread, ever: the fingerprint was the notification thread id, which never changes for a PR, so the first review request popped up and every later review, comment or merge on that PR was silent for at least 7 days. The fingerprint now includes the thread's updated_at (your own activity is still filtered by suppress_self)
• GitHub CI: after the first failing run on a PR, later failures on the same PR (fail → push → fail) never notified because the fingerprint was (PR, roll-up). It now includes the identity of the check runs (from gh pr checks --json link)
• github.ci_notify: ["pass"] was silently ignored — the app force-quieted every passing roll-up regardless of config. It now honours the quiet flag computed from ci_notify
• changing the theme or clicking Skip this version wrote the whole in-memory config back to disk. On first run that replaced the friendly template (with its _readme notes) by bare defaults; when config.json was unparsable it overwrote the user's real credentials with placeholders. Only the changed key is patched now, and nothing is written while the file cannot be parsed. DEFAULT_CONFIG is also deep-copied so those edits can no longer leak into the module defaults
• Windows: the two startup one-shot timers (first dependency check, first update check) call sender.stop(), but the Windows timer passed None as the sender, so both raised AttributeError on every fire — re-arming forever — and the startup check never ran (Status stuck on "Checking…", no "setup needed" hint, no update check). The Windows timer now passes itself as the sender, matching rumps, and a failing tick no longer kills the timer
• an unexpected exception inside the poll worker (e.g. a malformed gh payload) left the app stuck in "Checking…" — Check now greyed out and the spinner icon shown until relaunch — with the traceback lost on stderr. It is now logged, shown in Status ▸, and the UI state is restored
• the stale updater.__version__ (1.3.0) is now 1.5.8 so source runs report a sensible version
• Windows: the app crashed at startup (ValueError from pystray in set_menu). pystray only accepts menu callbacks with 0–2 positional parameters by __code__.co_argcount, and the adapter lambda carried two extra default-valued parameters. Callbacks are now built as two-argument closures; the fake_pystray test stub enforces the same rule so this cannot regress
• PagerDuty on-call status showed you as on-call when you were only the level-2/3 fallback on an escalation policy: /oncalls returns every level you sit on, and a schedule-less "direct target" entry took precedence over (and could never be replaced by) a real scheduled shift. Deeper levels are now listed with their own wording but do not count as on-call by default (pagerduty.oncall_max_level), and scheduled shifts win over direct targets regardless of the order PagerDuty returns them

Full compare: v1.5.8...v2.0.0-beta.1