No changes to the library code; this release is documentation and CI.

API docs layout fixed. The README’s wide banner image ran under the “On This Page” sidebar on the published docs. RDoc now loads assets/rdoc-overrides.css, which keeps images inside the content column.
Docs deploy on their own. The documentation workflow runs on every push to main that changes a file the docs render, and can be started by hand from the Actions tab on any branch. Release tags still publish the docs as before.
Faster CI for doc-only changes. Changes that touch only Markdown files or assets/ skip the test matrix. Tag pushes always run the full pipeline.
Docs checked on pull requests. A PR that changes a file the docs render now runs the RDoc build, so a broken build fails before merge. Nothing is deployed from a PR.
Workflow lint and wider test matrix. actionlint runs on the workflows, and the test matrix adds Ruby 3.4.
CI tidy-ups. Superseded pull request runs are cancelled. Documentation deploys queue instead of cancelling each other. Workflow permissions are narrowed, and the test job has a 20-minute timeout.

[v0.3.1] - 2026-10-22

No changes to the library code; this release is packaging and documentation.

Smaller gem. The package no longer includes assets/ (README images) or the unused lib/tasks/rdoc.rake: 52 KB instead of 973 KB.
API reference fixed. The published docs at dashapi-ai.github.io/cal-invite/ were missing lib/cal_invite.rb and every provider class (Google, Outlook, Office365, Yahoo, Ics, Ical, IcsContent, IcsDownload, BaseProvider), because the docs workflow’s lib/**/*.rb glob only matched one directory level. The workflow now runs rake rdoc and also publishes the README, CONFIGURATION.md, CACHING.md and CHANGELOG.md.
New CACHING.md on caching CalInvite output in your app (plain Ruby, Rails 8.1, Sinatra, Hanami, Grape), including HTTP caching for .ics feeds. The example app gains a cached .ics download (MeetingsController#show).
README rewritten for 0.3.0, with accurate examples that run in plain Ruby.
Gem metadata: the RubyGems changelog link pointed at a non-existent master branch; it now points at main, and a documentation link was added.

[v0.3.0] - 2026-10-22

Upgrading from 0.2.x

These changes can break existing callers:

Unknown Event attributes raise. Event.new and Event#update_attributes raise ArgumentError (Unknown attribute(s): locaton) for keys that aren’t Event attributes, instead of silently ignoring them. If you pass a whole params hash, slice it first. A rejected update_attributes leaves the event unchanged. (#10)
timezone is validated. It must be 'UTC' or an IANA identifier known to tzinfo (e.g. 'America/New_York'); anything else, including raw offsets like '+01:00', raises ArgumentError. Such values previously produced invalid .ics (a TZID= with no VTIMEZONE, times shifted by the offset) and could inject parameters. (#7)
Time order is validated. ArgumentError when end_time isn’t after start_time (all-day events may end on the same date), or a multi_day_sessions entry is missing a time or ends before it starts. (#9)
Provider names are validated. generate_calendar_url raises ArgumentError listing the supported providers for anything outside CalInvite::Providers::SUPPORTED_PROVIDERS (:ics_content still works as an alias of :ics). Previously e.g. :string resolved Ruby’s String and failed confusingly. (#11)
config.timezone is now honored as the default Event#timezone when timezone: isn’t passed, as documented. If you set it to something other than 'UTC', your events now render in that zone. An explicit timezone: still wins. (#22)
No more blanket active_support/core_ext. Requiring cal_invite no longer monkey-patches core classes (Integer#hours, Object#blank?, String#squish, …) in non-Rails apps. Plain-Ruby code that used those only because cal_invite loaded them must require "active_support/core_ext" (or the specific file). Rails apps are unaffected. This also lets the gem load with ActiveSupport 6.0 on Ruby 4.0. (#15)

Removed

Output caching. generate_calendar_url builds its output fresh on every call and never touches a cache store; update_attributes no longer wipes every event’s cache entries. The cache ignored cache_prefix, almost never hit, froze .ics DTSTAMP, and cost more than regenerating. Output is unchanged. To cache anyway, do it in your app; CONFIGURATION.md has an example. (#14)
Cache helpers. CalInvite.fetch_from_cache, write_to_cache, read_from_cache, clear_cache!, clear_event_cache!, clear_provider_cache! and CalInvite::Caching are gone. They had never worked (every call raised NoMethodError). CACHING.md is gone too. (#32)

Deprecated (removed in 0.4.0)

config.cache_store=, cache_prefix=, cache_expires_in= have no effect. They print warning: [cal_invite] Configuration#cache_store= is deprecated and has no effect ... pointing at your initializer line, but still store the value, so code that reads them keeps working. Delete them from your initializer. (#32)
::BaseProvider. The provider base class is now CalInvite::Providers::BaseProvider, so the gem no longer owns a top-level constant that can collide with your app’s. ::BaseProvider remains as an alias (not defined at all if your app has its own BaseProvider). Ruby prints constant ::BaseProvider is deprecated when deprecation warnings are enabled (e.g. RUBYOPT=-W:deprecated). (#13)

Security

Fix content-line/parameter injection in .ics output (:ics, :ical, IcsContent). A CR or LF in event attributes (title, description, attendee/organizer names and emails, uid, rrule, visibility, geo, …), or a " in a display name, could add arbitrary properties to a generated invite (e.g. extra attendees, a different METHOD:). URL providers were not affected. (#5)
TEXT values are escaped per RFC 5545 (backslashes were previously left unescaped and lone \r passed through).
CN= display names are quoted and escaped per RFC 6868.
URL: is no longer TEXT-escaped (it’s a URI; commas and semicolons in links were being corrupted).
Values that can’t be escaped raise ArgumentError if they contain a line break. visibility must be :public/:private/:confidential, geo numeric, and sequence an integer.
SECURITY.md: real policy (supported versions, private reporting via GitHub Security Advisories).

Added

organizer: accepts a plain email string (organizer: "jane@example.com"), like attendees; a string previously crashed .ics generation with TypeError. String-keyed hashes are read too. (#12)
Multi-day events no longer need top-level start_time/end_time: they default to the earliest session start and latest session end, so the README’s multi-day example works. :google/:outlook/:office365 show the overall span; :ics/:ical/:yahoo still emit one entry per session. Explicit times win. (#24)

Fixed

.ics content lines longer than 75 octets are folded (RFC 5545 §3.1), never splitting a multi-byte UTF-8 character. Long lines could be mangled or rejected by Outlook and mail gateways. (#6)
'UTC' events emit Z-suffixed DTSTART/DTEND instead of TZID=UTC with no VTIMEZONE. (#7)
:ical and IcsContent now produce exactly the same content as :ics. IcsContent previously emitted UTC-only times with no VTIMEZONE, ignored notes, and didn’t support all-day events. :ics all-day events without times default to today, as :ical did. (#4)
Non-UTC timezones under tzinfo 1.x (Rails/ActiveSupport 6.0): :ics/:ical raised NoMethodError (to_local) and silently omitted VTIMEZONE. (#17)
Event.new no longer mutates the attributes hash passed in (it used to delete several keys from it, so reusing the hash dropped them). (#8)

Docs

README “Testing” section covers rake test, bin/console, and the Example app’s local-path Gemfile workflow. CLAUDE.md merged into AGENTS.md.

[v0.2.1] - 2026-08-10

:google provider: honor Event#attendees/show_attendees (add=), Event#timezone (ctz=), Event#rrule (recur=), and Event#busy (crm=BUSY/AVAILABLE) in the generated Google Calendar URL — these were already modeled on Event for the ics providers but never wired into :google.
:outlook/:office365 providers: honor Event#busy via freebusy=.
:outlook/:office365 providers: fix stale deep-link URLs. Switched from the legacy outlook.live.com/calendar/0/action/compose and outlook.office.com/owa/ hosts to the currently documented outlook.live.com/calendar/deeplink/compose and outlook.office.com/calendar/deeplink/compose, and added the required rru=addevent param.
CI: fix Rails 6 + Ruby 3.3/3.4 test failures (missing logger/mutex_m/bigdecimal requires), fix RDoc doc generation (stale .rdoc_options crashing the darkfish generator and silently excluding all of lib/), and fix a GitHub Pages deploy deadlock (duplicate concurrency: group: "pages" between main.yml and the reusable documentation.yml).

[v0.2.0] - 2026-08-09

Add Event#organizer and a method: option (:publish/:request/:cancel) to generate_calendar_url, so ics/ical output can carry METHOD:REQUEST/METHOD:CANCEL + ORGANIZER/SEQUENCE/STATUS, matching what mail clients (Gmail, Outlook, Apple Mail) need to render an attached .ics as an RSVP-capable meeting invite (or a real cancellation) rather than a plain file
ATTENDEE lines now include CUTYPE/ROLE/PARTSTAT
IcsDownload.headers/.wrap_for_download accept method: to add the matching method=REQUEST Content-Type parameter
Fix a superclass mismatch crash when both the :ics provider and IcsDownload/IcsContent were loaded (dead duplicate Ics/Ical class definitions in ics_content.rb)
Fix: stable UID. Event#uid is now a real attribute (auto-generated and memoized per Event instance, or pass your own) instead of the ics/ical/ics_content providers each minting a fresh random UID on every generate call — required for mail/calendar clients to recognize a :request update or :cancel as referring to a previously sent invite rather than a new, unrelated event. See new Event#sequence (defaults to 0) and CONFIGURATION.md’s “Updating and cancelling invites”.
Fix: real VTIMEZONE. :ics/:ical now emit a complete VTIMEZONE component (STANDARD/DAYLIGHT observances with correct offsets and RRULEs, derived from TZInfo’s transition data) for any recognized IANA timezone, instead of an empty/absent block. DTSTART;TZID=.../DTEND;TZID=... are also now correctly converted from UTC to that zone’s local wall-clock time — previously the UTC clock digits were emitted verbatim under a non-UTC TZID, mislabeling the time. 'UTC' continues to use plain Z-suffixed timestamps with no VTIMEZONE, as before.
Add tzinfo as an explicit gem dependency (previously relied on it only transitively via activesupport).
Event#attendees entries can now be { email:, name:, partstat: } hashes (plain email strings still work) — ATTENDEE lines gain CN= and a settable PARTSTAT (:accepted/:declined/:tentative/:needs_action/:delegated) instead of always NEEDS-ACTION.
Add method: :reply, for building an attendee’s METHOD:REPLY back to an organizer (omits RSVP=TRUE, honors each attendee’s partstat:).
Add Event#geo (GEO: property), Event#reminders (VALARM blocks), Event#busy (TRANSP:), Event#visibility (CLASS:), Event#rrule (raw RRULE: recurrence value), and Event#calendar_name (X-WR-CALNAME on the VCALENDAR) — all optional, all emitted by :ics/:ical/IcsContent.
Add method: :counter/:decline_counter, generating METHOD:COUNTER/METHOD:DECLINECOUNTER for an attendee proposing a new time and an organizer rejecting it, respectively. (Note: :decline_counter correctly renders as METHOD:DECLINECOUNTER — one word per RFC 5545 — not a literal upcase of the Ruby symbol; fixed via a new BaseProvider#method_value used everywhere a METHOD:/method= value is emitted, including IcsDownload.headers.)
Add Event#importance (:low/:normal/:high → standard PRIORITY: + Outlook’s X-MICROSOFT-CDO-IMPORTANCE:) and Event#allow_counter (default true; false → Outlook’s X-MICROSOFT-DISALLOW-COUNTER:TRUE, hiding its “Propose New Time” action) and X-MICROSOFT-CDO-BUSYSTATUS alongside the existing TRANSP: — genuinely real, protocol-level, Outlook-specific properties (confirmed against Microsoft’s documented .ics extensions), safe to always emit since RFC 5545 requires unrecognized X- properties be ignored by other clients.
CONFIGURATION.md: rewrite “Guest permissions” — Google/Microsoft’s guest-permission toggles are confirmed API-only (no .ics property, no parameter on Google’s own “add to calendar” render URL either), but the section now documents the real Outlook-specific levers above as the closest addable equivalent. Rewrite “Attendee-proposed reschedules (COUNTER)” now that generation is implemented, with the caveat that mainstream mail clients still don’t reliably expose a “propose new time” UI to trigger one from. Add “Tracking RSVPs” (inbound METHOD:REPLY webhook vs. provider-API alternative, with the Gmail reliability caveat) for the RSVP-tracking gap, which remains genuinely outbound-only.
Example/calendar_app: add a working Meeting/MeetingAttendee persistence example (migration + models + MeetingsController#create/#reschedule/#cancel + MeetingMailer) demonstrating the uid/sequence lifecycle end-to-end, plus an illustrative EventRepliesController showing how to parse an inbound METHOD:REPLY email. The example app’s Gemfile still tracks the published gem — this demo needs 0.2.0, so run it locally against path: "../.." until that’s released.
Add per-attendee rsvp: override ({ email:, name:, partstat:, rsvp: }) so RSVP=TRUE can be suppressed even under method: :request — needed for a “registration confirmed” style invite where the recipient is already partstat: :accepted and nothing is actually being requested. Fixed a real bug found while adding it: the existing symbol/string dual-key lookup pattern (attendee[:key] || attendee["key"]) silently discards an explicit false value (false || x evaluates to x in Ruby, not false), which would have made rsvp: false a no-op; replaced with BaseProvider#attendee_hash_value, used for all attendee-hash reads.
CONFIGURATION.md: add “Replicating a ticketing-platform confirmation email” — a full worked example (matching a real Luma-style confirmation email byte-for-byte in structure, brand names genericized) covering the partstat: :accepted, rsvp: false attendee pattern, the full ActionMailer attachment setup, why From/Reply-To/ORGANIZER can legitimately all differ, and an explicit note that Apple Wallet .pkpass passes are an unrelated technology (PassKit, not iCalendar) and out of scope for this gem.