Breakage Calendar · Data for scripts and agents · Updated November 13, 2026

Breakage Calendar API

CyberNative AI LLC

Every date on the Breakage Calendar is in one JSON file: what stops working, who is affected, how to check, and the fix. A tested fix carries the day we ran it.

An end-of-life list says when a version stops getting updates. This calendar is about the day something stops working: an image tag that moves to a new major version, an AI model or API that shuts down, a CI runner that is retired, a time-zone rule that changes. It needs no key and no sign-up, so a coding agent can read it before it upgrades or deploys.

The files

Four static files on cybernative.ai. They change when we add or update a date.

URL	What it holds
`https://cybernative.ai/breakage-calendar/entries.json`	Every date: what changes, who is affected, how to check, the fix and its sources. The one source the page, the subscription and the feed are built from.
`https://cybernative.ai/breakage-calendar/check/signatures.json`	For each date a repository's files can show: which files to read and the search to run on each line. Dates a repository cannot show are listed with the reason.
`https://cybernative.ai/breakage-calendar/breakage-calendar.ics`	The same dates as a calendar subscription (iCalendar).
`https://cybernative.ai/breakage-calendar/feed.xml`	The same dates as an Atom feed, newest first.

Each date also has its own place on the page: https://cybernative.ai/breakage-calendar/#<id>, where id is the entry's id. Link there when you show a date to a person.

Two questions in one command

Which dates are still to come, with the kind of fix each has:

curl -s https://cybernative.ai/breakage-calendar/entries.json |
  jq -r --arg today "$(date -u +%F)" '.entries[]
    | select(.overdue or ((.dateEnd // .date) >= $today))
    | [.date, .fix.kind, .short, "https://cybernative.ai/breakage-calendar/#" + .id]
    | @tsv' | sort

Which dates name one dependency (here celery). It searches the entry's text and the libraries we found sending a retiring AI model by default:

curl -s https://cybernative.ai/breakage-calendar/entries.json |
  jq -r --arg dep celery '.entries[]
    | select([.title, .what, .affected, .check, (.checkCode // ""),
              ((.packages // [])[] | .package)]
             | join(" ") | ascii_downcase | test("\\b" + $dep + "\\b"))
    | "\(.date)  \(.short)  https://cybernative.ai/breakage-calendar/#\(.id)"'

A name search finds dates that mention the name. It cannot tell whether your version or your settings are affected: read the entry's affected and check fields, or run the repository check below.

Which dates hit a repository

signatures.json holds the search each date's own check runs. For each signature: read the files whose path matches files, and report the date when a line matches one of patterns.

Field	Meaning
`row`	The entry id in `entries.json`.
`files`	`any`: every text file. `source`: code and configuration, not documents. `workflow`: files matching the top-level `workflow` pattern (`.github/workflows/*.yml`). Anything else is a regular expression for the path, relative to the repository's top folder.
`patterns`	Regular expressions (JavaScript syntax), each tested against one line at a time.
`needs`	If present, a file counts only when its whole text also matches this expression (for example `ubuntu-latest`).
`kind`	`match`: a matching line uses what the date changes. `possible`: the files use it, and only some setups break; read the entry's check. `postgres`: a Docker image line whose tag moves; the reference matcher's `postgresVerdict` decides.
`blind`	Entry ids a repository's files cannot show (a server's certificate, a resolver's trust anchor), each with the reason.

The reference matcher is the scan function in match.js, JavaScript with no dependencies. It is the code the in-browser repository check runs; it skips .git, node_modules, virtual environments and build output, and lines in documents count only as mentions.

Fields in entries.json

The file is one object: title, page, updated (the day the file last changed) and entries, a list of dates. Each entry has:

Field	Meaning
`id`	Stable identifier: the date, then a short slug. Also the page anchor.
`date`	First day of the change, `YYYY-MM-DD`.
`datePrecision`	`day`; `month` (some day that month, `date` is the 1st); or `window` (from `date` to `dateEnd`).
`dateEnd`	Last day of a window, else `null`.
`days`	Optional. In a window, the separate days the change happens (brownouts), from `date` to `dateEnd`.
`time`	Optional time of day as text, for example `01:00 UTC` or `02:00 local`; else `null`.
`overdue`	Optional. The change was announced for `date` but had not happened on this day. Treat it as still to come.
`title`, `short`	A one-line title, and a shorter one for lists.
`area`	A broad group, for example `AI APIs`, `CI runners`, `Time zones`.
`what`	What changes.
`affected`	Who is affected, and who is not. Later updates in this text start with `Update, <day> <Month>:`.
`check`, `checkCode`	How to check your own system, and an optional command to run.
`fix.kind`	`tested`: we ran this fix (`fix.testedAt` is the day). `guide`: a check or guide; the fix itself is not tested by us. `none`: no tested fix yet.
`fix.href`, `fix.label`	Where the fix is. A path starting with `/` is on cybernative.ai.
`fix.text`, `fix.code`	Optional. The fix in words, and a command or snippet.
`sources`	The vendor's own announcements: `label`, `href`, and `checkedAt`, the day we last read it.
`packages`, `packagesNote`	Optional, on AI model retirements: libraries whose code sends the retiring model when you do not choose one, with the version and file line we read.
`added`	The day the entry joined the calendar.
`leadDays`	Optional. Days of warning the calendar subscription gives (default 14).
`see`	Optional. A related page on cybernative.ai: `href` and `label`.

Fields not listed here are for the page and the feed, and may change.

Fetching

No key and no account. Send a User-Agent that names your tool.
Each file has an ETag and Last-Modified: send If-None-Match and an unchanged file answers 304 with no body. With Accept-Encoding: gzip, entries.json is about 95 kB instead of 350 kB.
The files do not send CORS headers, so a web page on another site cannot read them from the visitor's browser. Fetch them from a server, a command line or an agent's own tools.
Something wrong or missing? Write to hello@cybernative.ai with the entry id.