Skip to content

Releases

Actions Monitor v2026.11.03

Choose a tag to compare

@github-actions github-actions released this 03 Nov 13:59
Immutable release. Only release title and notes can be modified.
4527547

2026-09-11

  • Releases are cut weekly — the scheduled build now runs once, on Tuesday morning (06:00 UTC), instead of three times a day. A release is still only created when main changed since the previous one, so quiet weeks stay quiet; the manual Release workflow is unchanged.

Actions Monitor v2026.09.22

Choose a tag to compare

@github-actions github-actions released this 22 Sep 11:21
Immutable release. Only release title and notes can be modified.
4bce2ca

2026-09-11

  • Releases are cut weekly — the scheduled build now runs once, on Tuesday morning (06:00 UTC), instead of three times a day. A release is still only created when main changed since the previous one, so quiet weeks stay quiet; the manual Release workflow is unchanged.

Actions Monitor v2026.09.15

Choose a tag to compare

@github-actions github-actions released this 15 Sep 11:31
Immutable release. Only release title and notes can be modified.
48c71a8

2026-09-11

  • Releases are cut weekly — the scheduled build now runs once, on Tuesday morning (06:00 UTC), instead of three times a day. A release is still only created when main changed since the previous one, so quiet weeks stay quiet; the manual Release workflow is unchanged.

Actions Monitor v2026.09.11

Choose a tag to compare

@github-actions github-actions released this 11 Sep 12:08
Immutable release. Only release title and notes can be modified.
929a00a

2026-07-15

  • Workflow form: no more dead space when resized taller — extra vertical space used to be distributed as odd empty gaps between the fixed-height fields (the form layout had no stretch sink). The config.yaml preview now takes stretch factor 1 (min height 150), so resizing the dialog grows the preview and everything else stays snug.

  • Settings: Edit crashed for quoted PR/URL entries — clicking Edit on a workflow whose config values were quoted (mode: "pr", mode: "url") raised yaml.representer.RepresenterError: cannot represent an object from the form's YAML preview: ruamel's round-trip loader (preserve_quotes=True) returns ScalarString subclasses that yaml.safe_dump refuses, and the form re-inserted them into the rebuilt entry. New to_plain() recursively converts ruamel container/scalar types to plain Python; applied when an entry enters WorkflowFormDialog and defensively inside entry_to_yaml(). Regression test covers editing ruamel-loaded pr and url entries.

  • Sort menu polish (user feedback) — the ▲/▼ glyphs in the sort dropdown and active sort label carry U+FE0E (text presentation selector) so Windows renders them monochrome instead of hijacking them with the colour-emoji font, and the menu now pops up anchored under the section's Sort label (right-aligned) instead of floating at the cursor, which could land it detached beyond the window edge.

  • Settings polish (user feedback) — notification-type rows gained "Sound" / "Enabled" column headers (plus tooltips on the dropdown/checkbox) so it's clear what each control does; labels inside cards no longer render dark background boxes (QLabel { background: transparent } reset at dialog level, same root cause as the row-card patches); long labels (staleness note, token how-to steps, notification subtitles, empty-list hint) now word-wrap instead of forcing pages wider than the dialog; the add/edit-workflow dialog opens tall enough (up to 940px, clamped to screen height) that all fields plus Cancel / Add workflow are visible without scrolling.

  • Row cards paint uniformly — the app-wide QWidget { background-color } stylesheet rule painted BG_DARK patches behind the row's child container widgets (badge row, right column, name/branch spacer) on top of the v2 card fill, making row backgrounds look blotchy. WorkflowRow stylesheets now append WorkflowRow QWidget { background-color: transparent } so the card fill (and hover/snooze/blink variants) shows through everywhere; widget-level styles (accent bar, badges, snooze hover) still win. Blink-on-focus goes through a new paint_bg() helper so mid-blink frames keep the transparency reset too.

  • ACTIONS_MONITOR_BETA env override + task run:beta — beta flags can be force-enabled per process via a comma-separated env var (e.g. ACTIONS_MONITOR_BETA=settings_ui) without editing config.yaml; task run:beta wraps it for source runs. The override copies the merged beta dict before mutating — _deep_merge shallow-copies only one level, so writing through the shared reference would have poisoned DEFAULT_CONFIG for every later reload (caught by the test suite).

  • Integration test suite + screenshot harness — new tests/ (33 tests) driving the real Qt UI offscreen with no network: poller threads are kept dormant and poll results injected as StatusEvents (app_env fixture with per-test temp config/state). Covers empty state, row creation/removal, summary-strip counts, section counts, snooze (incl. summary exclusion), sort menu persistence, red failure info lines, error rows, beta-flag gating (incl. hot-reload), settings pages writing config.yaml (comment preservation, invalid-duration/regex/YAML rejection), workflow form mode detection + entry building. test_screenshot_* tests save PNGs of the main window and every settings page to artifacts/screenshots/ (gitignored). New Taskfile.yml (task install-dev / test / screenshots / run / build). Screenshot pass caught and fixed two real layout bugs: the settings content pane rendered on the default light palette without the app stylesheet (tests now apply it for parity), and non-wrapping field labels forced the add-workflow form wider than its dialog (labels now word-wrap; form min width 660).

  • In-app Settings window (beta) — new src/settings_ui.py implementing the Settings Window design: left-nav dialog with five pages. Workflows: list of configured entries (name, mode badge, summary, polling rate) with Add/Edit/Remove; the add/edit form auto-detects the mode from a pasted GitHub URL (workflow file → branch/PR, ?query=actor: → actor, search/pulls URL → Search URL with query extraction), shows per-mode option cards (branch filter; PR max rows / stale-after / extra + ignore workflows; actor failed-filter; search query) and a live YAML preview of the entry. Notifications: per-type enable + sound, delivery (batch window / suppress-older-than / toast duration), PR-mode overrides with tri-state inherit. PR rules: staleness thresholds, Jira base URL, bot-reviewer regex with presets. GitHub account: token field + async "Test token" (login + rate-limit remaining). Raw YAML: in-dialog editor with YAML validation on save. Writes go to config.yaml immediately via ruamel.yaml round-trip (user comments preserved; PyYAML fallback) — new ruamel.yaml dependency — and reach the pollers through the existing hot-reload watcher. Opened from the beta-gated header gear / footer "⚙ Settings" link; with the flag off those entry points fall back to opening config.yaml in an editor.

  • Main window refreshed to design v2 (design/actions-monitor-improvement-designs/) — rows are now uniform rounded cards (#262220, 6px radius, 3px gap, hover #2C2825) with an inset pill accent bar, replacing zebra striping; snoozed rows dim to #221F1D. New summary strip under the header shows a worst-status dot, N failed · N running · N passing counts, and an Updated HH:MM timestamp. Header is 52px with 30×30 hover-rounded icon buttons (18px glyphs) and a 30px logo. Section headers gained a row count and an inline Sort ▾ dropdown menu (replaces the per-section sort bar; same six modes + clear, state.json persistence unchanged). Footer collapsed to a single row (#232020): toggles left, config/settings link right. First-run empty state with an "Add a workflow" call-to-action. Failed/errored rows render their status line in red. Badges bumped to 10px/2×6px padding, titles to 13px #F5F5F4, status icons to 22px. New _make_settings_icon() gear in icons.py.

  • beta.settings_ui feature flag — new beta: config block (all switches off by default, documented in config.template.yaml). When settings_ui: true, the header shows a gear button (after a divider) and the footer link becomes "⚙ Settings", both opening the upcoming in-app Settings window; when off, the footer keeps the "Open config ↗" editor link and the gear stays hidden. Gating is re-applied on config hot-reload via _apply_beta_flags().

  • "Start with Windows" self-heals stale registry entries — the HKCU Run entry survives repo moves and source→exe migrations pointing at a command that no longer exists (observed: an old pythonw.exe C:\Repos\...\main.py source path), so startup silently launched nothing while the checkbox still showed enabled. Frozen builds now call StartupManager.heal() at launch: if an entry exists and differs from the current "{sys.executable}" command, it is rewritten. Source runs never touch the entry (so dev runs don't hijack an exe install's startup).

  • PR rows no longer show grey "Unknown" for real build states (#16) — three fixes. (1) _worst_status() now ranks all seven statuses (failure > running > queued > cancelled > success > skipped > unknown) via _STATUS_PRIORITY instead of falling through to ST_UNKNOWN when a PR's runs are all cancelled/skipped — a branch whose only in-window run was a skipped E2E sibling used to render a grey question mark. (2) Per-branch fill-in: PRWorkflowPoller._poll() now fetches the latest run of every configured workflow the bulk actor-filtered fetch didn't surface for an open-PR branch (runs by other actors, or older than the recent window) instead of only branches with zero runs — cheap via per_page=1 + the 30s TTL / ETag cache layers. (3) Head-SHA fallback: a PR with no runs from any configured workflow at all (e.g. an acceptance → production PR whose CI lives in production workflows) now falls back to the latest run on its head SHA (fetch_runs_by_sha, mirroring URL mode) so the row reports the PR's real CI status; _cache_pr additionally stores head_sha. Also added startup_failure (→ failure) and stale (→ cancelled) to CONCLUSION_MAP, and waiting/pending/requested API statuses now resolve to queued instead of unknown.

  • Section sorts are now independent per section (#17) — _cycle_sort() used to clear every other section's sort and re-order those sections back to default whenever any sort label was clicked. Each section now keeps its own Status/Updated/Created sort; state.json persistence already supported multiple entries and is unchanged.

  • Full code review sweep — 20+ fixes across all modules. Highlights:

    • pollers/gh_api: _fetch_user_open_prs dropped the nonexistent creator= param (the Pulls list API ignores it) and now fetches 100 most-recently-updated open PRs with client-side author filtering — previously a busy repo could push the user's PRs out of the 50-item window, silently removing their rows. fetch_actor_runs conclusion filter now uses the runs API's status= param (a conclusion= param is ignored), so actor mode filter: "failed" actually filters server-side. A run that starts and finishes between two polls now fires its success/failure outcome instead of a misleading "started" toast that permanently swallowed the completion notification (both branch/actor `_detect_notifica...
Read more

Actions Monitor v2026.07.16

Choose a tag to compare

@github-actions github-actions released this 16 Jul 07:17
Immutable release. Only release title and notes can be modified.

2026-07-15

  • Workflow form: no more dead space when resized taller — extra vertical space used to be distributed as odd empty gaps between the fixed-height fields (the form layout had no stretch sink). The config.yaml preview now takes stretch factor 1 (min height 150), so resizing the dialog grows the preview and everything else stays snug.

  • Settings: Edit crashed for quoted PR/URL entries — clicking Edit on a workflow whose config values were quoted (mode: "pr", mode: "url") raised yaml.representer.RepresenterError: cannot represent an object from the form's YAML preview: ruamel's round-trip loader (preserve_quotes=True) returns ScalarString subclasses that yaml.safe_dump refuses, and the form re-inserted them into the rebuilt entry. New to_plain() recursively converts ruamel container/scalar types to plain Python; applied when an entry enters WorkflowFormDialog and defensively inside entry_to_yaml(). Regression test covers editing ruamel-loaded pr and url entries.

  • Sort menu polish (user feedback) — the ▲/▼ glyphs in the sort dropdown and active sort label carry U+FE0E (text presentation selector) so Windows renders them monochrome instead of hijacking them with the colour-emoji font, and the menu now pops up anchored under the section's Sort label (right-aligned) instead of floating at the cursor, which could land it detached beyond the window edge.

  • Settings polish (user feedback) — notification-type rows gained "Sound" / "Enabled" column headers (plus tooltips on the dropdown/checkbox) so it's clear what each control does; labels inside cards no longer render dark background boxes (QLabel { background: transparent } reset at dialog level, same root cause as the row-card patches); long labels (staleness note, token how-to steps, notification subtitles, empty-list hint) now word-wrap instead of forcing pages wider than the dialog; the add/edit-workflow dialog opens tall enough (up to 940px, clamped to screen height) that all fields plus Cancel / Add workflow are visible without scrolling.

  • Row cards paint uniformly — the app-wide QWidget { background-color } stylesheet rule painted BG_DARK patches behind the row's child container widgets (badge row, right column, name/branch spacer) on top of the v2 card fill, making row backgrounds look blotchy. WorkflowRow stylesheets now append WorkflowRow QWidget { background-color: transparent } so the card fill (and hover/snooze/blink variants) shows through everywhere; widget-level styles (accent bar, badges, snooze hover) still win. Blink-on-focus goes through a new paint_bg() helper so mid-blink frames keep the transparency reset too.

  • ACTIONS_MONITOR_BETA env override + task run:beta — beta flags can be force-enabled per process via a comma-separated env var (e.g. ACTIONS_MONITOR_BETA=settings_ui) without editing config.yaml; task run:beta wraps it for source runs. The override copies the merged beta dict before mutating — _deep_merge shallow-copies only one level, so writing through the shared reference would have poisoned DEFAULT_CONFIG for every later reload (caught by the test suite).

  • Integration test suite + screenshot harness — new tests/ (33 tests) driving the real Qt UI offscreen with no network: poller threads are kept dormant and poll results injected as StatusEvents (app_env fixture with per-test temp config/state). Covers empty state, row creation/removal, summary-strip counts, section counts, snooze (incl. summary exclusion), sort menu persistence, red failure info lines, error rows, beta-flag gating (incl. hot-reload), settings pages writing config.yaml (comment preservation, invalid-duration/regex/YAML rejection), workflow form mode detection + entry building. test_screenshot_* tests save PNGs of the main window and every settings page to artifacts/screenshots/ (gitignored). New Taskfile.yml (task install-dev / test / screenshots / run / build). Screenshot pass caught and fixed two real layout bugs: the settings content pane rendered on the default light palette without the app stylesheet (tests now apply it for parity), and non-wrapping field labels forced the add-workflow form wider than its dialog (labels now word-wrap; form min width 660).

  • In-app Settings window (beta) — new src/settings_ui.py implementing the Settings Window design: left-nav dialog with five pages. Workflows: list of configured entries (name, mode badge, summary, polling rate) with Add/Edit/Remove; the add/edit form auto-detects the mode from a pasted GitHub URL (workflow file → branch/PR, ?query=actor: → actor, search/pulls URL → Search URL with query extraction), shows per-mode option cards (branch filter; PR max rows / stale-after / extra + ignore workflows; actor failed-filter; search query) and a live YAML preview of the entry. Notifications: per-type enable + sound, delivery (batch window / suppress-older-than / toast duration), PR-mode overrides with tri-state inherit. PR rules: staleness thresholds, Jira base URL, bot-reviewer regex with presets. GitHub account: token field + async "Test token" (login + rate-limit remaining). Raw YAML: in-dialog editor with YAML validation on save. Writes go to config.yaml immediately via ruamel.yaml round-trip (user comments preserved; PyYAML fallback) — new ruamel.yaml dependency — and reach the pollers through the existing hot-reload watcher. Opened from the beta-gated header gear / footer "⚙ Settings" link; with the flag off those entry points fall back to opening config.yaml in an editor.

  • Main window refreshed to design v2 (design/actions-monitor-improvement-designs/) — rows are now uniform rounded cards (#262220, 6px radius, 3px gap, hover #2C2825) with an inset pill accent bar, replacing zebra striping; snoozed rows dim to #221F1D. New summary strip under the header shows a worst-status dot, N failed · N running · N passing counts, and an Updated HH:MM timestamp. Header is 52px with 30×30 hover-rounded icon buttons (18px glyphs) and a 30px logo. Section headers gained a row count and an inline Sort ▾ dropdown menu (replaces the per-section sort bar; same six modes + clear, state.json persistence unchanged). Footer collapsed to a single row (#232020): toggles left, config/settings link right. First-run empty state with an "Add a workflow" call-to-action. Failed/errored rows render their status line in red. Badges bumped to 10px/2×6px padding, titles to 13px #F5F5F4, status icons to 22px. New _make_settings_icon() gear in icons.py.

  • beta.settings_ui feature flag — new beta: config block (all switches off by default, documented in config.template.yaml). When settings_ui: true, the header shows a gear button (after a divider) and the footer link becomes "⚙ Settings", both opening the upcoming in-app Settings window; when off, the footer keeps the "Open config ↗" editor link and the gear stays hidden. Gating is re-applied on config hot-reload via _apply_beta_flags().

  • "Start with Windows" self-heals stale registry entries — the HKCU Run entry survives repo moves and source→exe migrations pointing at a command that no longer exists (observed: an old pythonw.exe C:\Repos\...\main.py source path), so startup silently launched nothing while the checkbox still showed enabled. Frozen builds now call StartupManager.heal() at launch: if an entry exists and differs from the current "{sys.executable}" command, it is rewritten. Source runs never touch the entry (so dev runs don't hijack an exe install's startup).

  • PR rows no longer show grey "Unknown" for real build states (#16) — three fixes. (1) _worst_status() now ranks all seven statuses (failure > running > queued > cancelled > success > skipped > unknown) via _STATUS_PRIORITY instead of falling through to ST_UNKNOWN when a PR's runs are all cancelled/skipped — a branch whose only in-window run was a skipped E2E sibling used to render a grey question mark. (2) Per-branch fill-in: PRWorkflowPoller._poll() now fetches the latest run of every configured workflow the bulk actor-filtered fetch didn't surface for an open-PR branch (runs by other actors, or older than the recent window) instead of only branches with zero runs — cheap via per_page=1 + the 30s TTL / ETag cache layers. (3) Head-SHA fallback: a PR with no runs from any configured workflow at all (e.g. an acceptance → production PR whose CI lives in production workflows) now falls back to the latest run on its head SHA (fetch_runs_by_sha, mirroring URL mode) so the row reports the PR's real CI status; _cache_pr additionally stores head_sha. Also added startup_failure (→ failure) and stale (→ cancelled) to CONCLUSION_MAP, and waiting/pending/requested API statuses now resolve to queued instead of unknown.

  • Section sorts are now independent per section (#17) — _cycle_sort() used to clear every other section's sort and re-order those sections back to default whenever any sort label was clicked. Each section now keeps its own Status/Updated/Created sort; state.json persistence already supported multiple entries and is unchanged.

  • Full code review sweep — 20+ fixes across all modules. Highlights:

    • pollers/gh_api: _fetch_user_open_prs dropped the nonexistent creator= param (the Pulls list API ignores it) and now fetches 100 most-recently-updated open PRs with client-side author filtering — previously a busy repo could push the user's PRs out of the 50-item window, silently removing their rows. fetch_actor_runs conclusion filter now uses the runs API's status= param (a conclusion= param is ignored), so actor mode filter: "failed" actually filters server-side. A run that starts and finishes between two polls now fires its success/failure outcome instead of a misleading "started" toast that permanently swallowed the completion notification (both branch/actor `_detect_notifica...
Read more

Actions Monitor v2026.05.16

Choose a tag to compare

@github-actions github-actions released this 16 May 07:11
Immutable release. Only release title and notes can be modified.

2026-05-15

  • Daily Build retries on transient runner starvation — today's scheduled run (25906763857) failed because the release / check job sat in queued for ~23 minutes without ever being assigned an ubuntu-latest runner; GitHub eventually gave up, dependents were skipped, and the run was marked failure. build.yml now schedules two follow-up crons at 06:00 and 07:00 UTC (in addition to the original 05:00). The _release.yml check job is already idempotent — once HEAD equals the latest release's target_commitish, should_build=false short-circuits the build/release/scoop jobs — so a successful first run turns subsequent retries into ~10-second no-ops. No retry-state file or external coordination needed.

Actions Monitor v2026.05.13

Choose a tag to compare

@github-actions github-actions released this 13 May 07:47
Immutable release. Only release title and notes can be modified.

2026-05-12

  • ignore_workflows opt-out (PR mode) — sibling of extra_workflows for silencing small / shadow-IT workflows that piggyback on a PR. Two independent switches per entry: status (drop run from row colour + representative-run pick) and notifications (drop run from notification transition tracking). Short form (- some-check.yml) ignores both; granular form takes a dict (file: plus optional status: / notifications:, default true each). PRWorkflowPoller.__init__ parses cfg_entry["ignore_workflows"] into two sets via _parse_ignore_workflows(); _poll() partitions each PR's group_runs into status_runs + notif_runs (matched by workflow-file basename), drives agg_status / representative_run / snoozed-row aggregate from status_runs, and tracks cur_run_ids + notif_agg for transition detection from notif_runs. Stored _prev_statuses[sub_key] now tracks the notif aggregate (which may diverge from the displayed row status when the two ignore sets differ). Documented next to extra_workflows in config.template.yaml.

  • Release notes shown in update dialog — UpdateDialog now renders the GitHub release body markdown via a QTextBrowser placed between the version/link block and the managed-cmd / status block. Source is UpdateChecker._release_data["body"] (already cached by _check_release(), no extra API call). Uses Qt's built-in CommonMark renderer (setMarkdown), dark-theme QSS (BG_ROW bg, FG_TEXT fg, FG_LINK for a), 150px fixed height with vertical scroll, external links open in browser. When the release body is empty the widget is hidden and the dialog stays its original 420×260/280 size; when present it grows to 480 wide and adds 170px height. Both install paths (direct Update/Skip + managed-install Copy/Close) show notes. Users can now see what's changing before clicking Update.

  • Replaced em-dash with hyphen across all UI-visible strings — WorkflowRow._update_labels status-info line (Success — run #9638 → Success - run #9638), snooze tooltips, UpdateDialog window title + completion label, URLQueryPoller "Cannot resolve @me" error, gh_api rate-limit + network error strings shown in row info lines, tray tooltip ({APP_NAME} — Success → {APP_NAME} - Success), and missing-deps / missing-libraries dialog titles. Comments and docstrings still use em-dashes. Avoids AI-stylised punctuation in user-facing copy.

Actions Monitor v2026.05.08

Choose a tag to compare

@github-actions github-actions released this 08 May 13:29
Immutable release. Only release title and notes can be modified.

2026-05-08

  • Richer notification content + multi-event toast aggregation — single notifications now include PR title, branch (with prefix), → target for PR mode, and the Jira key parsed from the branch — previously the body was just {workflow_name} • Run #{n}. Title now embeds workflow name + verb (✗ Acceptance failed). Batched toasts (when multiple notifications coalesce inside notifications.batch_window) now render a header summary line (✗ 2 failed • ▶ 1 started) followed by per-event lines (up to 4, then … and N more), sorted failure → new_run → success. Added a line field on _PendingNotification carrying the single-line summary; WorkflowPoller._build_notification(notif_type, state, is_pr) produces (title, body, line) from existing WorkflowState fields — no extra API calls. Toast-click URL falls back to state.pr_url when state.run_url isn't set.

  • Codebase audit cleanup — extracted status constants (ST_*, CONCLUSION_MAP, _STATUS_PRIORITY, _resolve_status) into a new pure-data src/status.py module, eliminating the triple definition that previously lived in pollers.py, icons.py (as _ST_*), and the doc claim that they lived in main.py. pollers.py re-exports for backward compat so existing call sites (from pollers import ST_*) keep working. Reordered MainWindow._apply_event to call row.update() before committing self._states[key] — a row.update() exception used to leave the state map ahead of the UI. Tightened the new_run notification guard in PRWorkflowPoller._poll to fire only when a previously-unseen run ID appears (pure GitHub run-history shrinkage no longer spams). Replaced three silent except Exception: pass blocks in pollers.py (_fetch_user_open_prs, extra-workflow _fetch_branch_runs) and one in gh_api._cached_unresolved_fetch with print(..., file=sys.stderr) calls so token / network failures are diagnosable from the console.

  • Per-section include_in_tray_status + notifications_enabled flags — needed because URL mode surfaces PRs the user doesn't own (e.g. an "Open for Review" inbox); failures there shouldn't turn the tray red and shouldn't notify. Added pollers.section_flags(cfg_entry) returning both flags with mode-aware defaults: include_in_tray_status=true for every mode; notifications_enabled=true for branch/pr/actor, false for URL mode. MainWindow._update_tray() now filters out states from sections with include_in_tray_status: false before calling _combined_status(). WorkflowPoller._fire_notification() early-returns when the section flag is off, replacing URL mode's previous "never call _fire_notification" pattern with a single uniform gate. Documented under "Section flags" in CLAUDE.md and on the URL example in config.template.yaml.

Actions Monitor v2026.05.07

Choose a tag to compare

@github-actions github-actions released this 07 May 07:39
Immutable release. Only release title and notes can be modified.

2026-05-06

  • Daily Build no longer self-perpetuates via Scoop manifest bump — _release.yml's check job compared git rev-parse HEAD against the latest release's targetCommitish, but the update-scoop job pushes bucket/actionsmonitor.json to main after every release. Result: each Daily Build saw "main moved" (the bot's scoop bump from yesterday), built a fresh release, and the cycle repeated forever — releases page showed a new tag every morning even when no human had committed. The check step now runs git diff --name-only LATEST HEAD -- . ':(exclude)bucket/actionsmonitor.json'; if the only change since the last release is the scoop manifest, should_build=false and the run skips cleanly. Manual workflow_dispatch of Release is unaffected because real commits land outside the exclude path.

Actions Monitor v2026.05.06

Choose a tag to compare

@github-actions github-actions released this 06 May 07:34
Immutable release. Only release title and notes can be modified.

2026-05-01

  • Relicensed under Business Source License 1.1 (BUSL-1.1) — replaced the bespoke WizX20 Free Use License with the SPDX-listed BUSL-1.1 (Licensor: WizX20; Change Date: 2030-05-01; Change License: Apache 2.0). The Additional Use Grant preserves the original intent — free for personal, internal, organisational, academic, and free-redistribution use — while explicitly forbidding sale, paid sublicensing, rental, and paid hosted/embedded distribution. NOTICE retention requirement is restated inside the Additional Use Grant. After the Change Date, each release auto-converts to Apache 2.0. Updated LICENSE, NOTICE, README "License" section, CONTRIBUTING.md contributor agreement, DEVGUIDE.md winget submission table, and the Scoop manifest's license field.

  • Split src/main.py into pollers.py + widgets.py + notifications.py — extracted the four WorkflowPoller variants (~1,090 lines) into src/pollers.py along with the WorkflowState / StatusEvent dataclasses, status constants, snooze registry, and the cross-poller helpers (_resolve_status, _deep_merge, parse_branch_prefix, extract_jira_key, parse_duration, _format_age, _worst_status, _combined_status). Pulled WorkflowRow + label/badge helpers (~470 lines) into src/widgets.py along with the colour palette (COLOUR, STATUS_LABEL, BG_*, FG_*). Moved NotificationManager, _PendingNotification, _NAMED_SOUNDS, _find_linux_default_sound, _ensure_focus_vbs, and the NOTIF singleton into src/notifications.py. main.py shrank from 4,272 → 2,217 lines. PyInstaller-safe: same one-way-import pattern as update.py — no extracted module imports from main. The snooze registry mutations move through new pollers.add_snooze / discard_snooze / clear_snooze helpers so MainWindow no longer touches the lock directly. notifications.configure(app_name=..., app_ico=..., focus_vbs=..., focus_signal=...) runs once at module init for the per-process paths. Smoke-tested in source mode.

  • Extract update flow into src/update.py — pulled out UpdateChecker, UpdateDialog, _detect_install_source, _MANAGED_UPGRADE_CMD, and _cleanup_stale_mei_dirs. main.py shrank from 4867 → 4272 lines; update.py is 687 lines. PyInstaller-safe: a naïve from main import ... in update.py triggered a circular reload at runtime because PyInstaller executes the entry script as both __main__ and main. Fixed by making update.py self-contained — it declares module-level placeholders for the constants + widget class it needs and exposes a configure(**kwargs) injection function. main.py calls update.configure(app_name=APP_NAME, build_commit=BUILD_COMMIT, is_windows=IS_WINDOWS, fg_text=FG_TEXT, ..., clickable_label_cls=_ClickableLabel) once at module init, after all those names are defined. Smoke-tested in source mode and as a fresh --onedir build.

  • Extract GitHub API plumbing into src/gh_api.py — pulled out the HTTP layer, rate-limit gate, ETag conditional-request cache, GraphQL helper, retry wrapper, friendly-error mapper, URL parsers/builders (parse_workflow_url, _build_workflow_url, _build_branch_url, parse_actor_url), endpoint fetchers (fetch_latest_run, fetch_pr_runs, fetch_actor_runs, fetch_github_username), the username cache and its public reset_username_cache() hook, the generic _prune_cache helper + _PR_CACHE_MAX / _REVIEW_CACHE_MAX size caps, and the PR-review aggregation pipeline (_compile_bot_regex, _aggregate_review_status, _cached_review_fetch, _cached_unresolved_fetch, _UNRESOLVED_THREADS_QUERY). main.py shrank from 5340 → 4867 lines; gh_api.py is 526 lines. Self-contained — no main imports — so all module-level state (etag cache, rate-limit cooldown, username cache) lives in one place and is shared across pollers via normal module imports rather than global. _reload_pollers now calls gh_api.reset_username_cache() instead of mutating the global directly. _resolve_status stays in main.py because it depends on the ST_* / CONCLUSION_MAP constants. Smoke-tested as source and as a fresh --onedir build.

  • Extract icon rendering into src/icons.py — pulled all PIL drawing code out of main.py into a new self-contained module: Lucide-inspired status glyphs (_draw_lucide_*), header icons (refresh / update / help), reviewer (user / bot) glyphs, snooze bell, embedded WizX20 mark + decoder, app/tray base icon composition, the _pil_to_qpixmap Qt bridge, the _status_qpixmaps / _snooze_qpixmaps / _base_icon_cache / _wizx20_mark_cache / _REVIEWER_ICON_B64 caches, and the _init_* helpers. main.py shrank from 6486 → 5340 lines; icons.py is 1202 lines. icons.py owns its own copy of COLOUR_BG, the seven status string literals, and the amber _FG_LINK default — the only state shared is the cache dicts which main imports directly. _generate_app_ico and _generate_check_glyph now take the destination Path as an argument instead of reading module globals; updated call sites in main.py, src/build.bat, src/build.sh, and .github/workflows/_build.yml. The icons import sits below the Pillow dep-check block so missing-Pillow still surfaces the friendly Qt dialog rather than a bare ImportError. Smoke-tested both as source and as a fresh --onedir build — every public icon function still renders and MainWindow constructs cleanly.

  • Switched from PyInstaller --onefile to --onedir + zip distribution — the previous onefile build extracted the full Python runtime (including python312.dll, vcruntime140*, the api-ms-win-crt-* set, and dozens of .pyd modules) to %TEMP%\_MEI<pid>\ on every launch. After a self-update Defender was racing the bootstrap and quarantining individual DLLs mid-extraction, leaving the new exe to die with Failed to load Python DLL '...\python312.dll'. LoadLibrary: The specified module could not be found. (the classic transitively-missing-dep form of that error). Onedir leaves the runtime files on disk in _internal/ next to the exe permanently, so AV scans them once at install/update time and never has a chance to race the loader. Release assets are now ActionsMonitor.zip / ActionsMonitor-linux.zip instead of bare binaries, each wrapping a top-level ActionsMonitor/ (or ActionsMonitor-linux/) folder containing the exe plus _internal/. CI build (_build.yml), local build (src/build.bat), Scoop manifest (bucket/actionsmonitor.json — adds extract_dir: "ActionsMonitor"), and release publishing (_release.yml) all flow through the new layout. Existing installs need a one-time manual reinstall — auto-update from a pre-2026-05-01 binary won't carry over because the old UpdateChecker only knows how to fetch and swap a single .exe. Grab the new zip from the release page, unzip, and run ActionsMonitor.exe from inside the folder.

  • UpdateChecker rewritten for the zip + folder swap — _apply_release_update now downloads the release zip to <install_dir>/.am_update.zip, verifies size + sha256, extracts to <install_dir>/.am_update_staging/, locates the wrapper folder by searching for the exe (robust to layout tweaks), and stores the wrapper path on _update_path. restart_app's helper batch/shell now does a four-step swap with rollback at every stage: rename _internal → _internal.old, rename <exe> → <exe>.old, move new exe in, move new _internal in. Any failure restores all backups before exiting non-zero. Post-swap the helper deletes the .old backups and the staging dir, then runs the same AV-warmup read across the new exe and every *.dll/*.pyd in _internal/ before launching — onefile only needed to warm the wrapper exe; onedir has the full runtime sitting on disk that AV needs to clear first. Startup leftover-cleanup (main()) extended to wipe legacy .old/.update sidecars, the new <exe>.old + _internal.old backups, .am_update.zip, and .am_update_staging/ so a force-killed helper can't leave the install dir half-swapped.

  • config.template.yaml lookup falls back to _MEIPASS — onedir bundles --add-data files into _internal/, which sys._MEIPASS points at. _APP_DIR / "config.template.yaml" (next to exe) won't exist in onedir installs, so ConfigManager._write_default() would fall through to the bare DEFAULT_CONFIG and ship users a config without the documented examples / per-mode comments. New _bundled(name) helper checks _APP_DIR first (preserves user-overridable copy + scoop's persist model), then _MEIPASS, returning the first match. Used for the template lookup; ready to drop in for any future bundled assets.