Crosspost
Ledger
Six vendors. One feed. Capability adds, breaking changes, deprecations, pricing.
Source of truth: joshduffy.dev/ledger. Updated hourly via hash-skip ingest.
-
Durable Objects new apac-ne and apac-se location hints
Two new locationHint values — apac-ne (Northeast Asia-Pacific) and apac-se (Southeast Asia-Pacific) — are now available for Durable Objects, allowing finer-grained placement within Asia-Pacific beyond the existing apac hint.
-
Durable Objects stay alive during outbound connections
Durable Objects now remain alive for the duration of active outbound TCP/WebSocket connections (up to 15 minutes), preventing premature eviction mid-stream. This directly benefits agents and LLM streaming patterns using connect() or outbound WebSockets.
-
Codex 0.142.0-alpha series released (alpha.1–alpha.6) 0.142.0-alpha.6
Six alpha pre-releases of Codex 0.142.0 have shipped in rapid succession between June 18–19, 2026. No detailed changelogs are published yet; treat as early-access / unstable.
-
Codex Rust v0.142.0-alpha.6 released rust-v0.142.0-alpha.6
Alpha pre-release 0.142.0-alpha.6 of the Rust Codex CLI is available. No detailed release notes provided; tracks ongoing alpha development.
-
Durable Objects: new apac-ne and apac-se location hints
Two new `locationHint` values (`apac-ne` for Northeast Asia-Pacific, `apac-se` for Southeast Asia-Pacific) are now available for Durable Objects, enabling finer-grained placement within the Asia-Pacific region.
-
Durable Objects stay alive during outbound connections
Durable Objects now remain alive for the duration of active outbound TCP or WebSocket connections (up to 15 min), preventing mid-stream eviction. This is critical for agents/Workers streaming LLM responses over outbound connections.
-
Durable Objects: New apac-ne and apac-se location hints
Durable Objects now supports two new location hints, `apac-ne` (Northeast Asia-Pacific) and `apac-se` (Southeast Asia-Pacific), for finer-grained placement within the Asia-Pacific region. Use these instead of the broader `apac` hint when traffic is concentrated in one sub-region.
-
Durable Objects: New apac-ne and apac-se location hints
Durable Objects now supports two new finer-grained location hints—`apac-ne` (Northeast Asia-Pacific) and `apac-se` (Southeast Asia-Pacific)—in addition to the existing broad `apac` hint. Useful for reducing latency when traffic is concentrated in one Asia-Pacific sub-region.
-
create-astro adds AGENTS.md and CLAUDE.md to new projects create-astro@5.1.0-beta.0
New Astro projects scaffolded with `create-astro` now include a default `AGENTS.md` file with dev server instructions, plus a `CLAUDE.md` symlink pointing to it. This improves out-of-the-box AI agent support for new projects.
-
@astrojs/cloudflare speeds up astro sync by skipping runtime 14.0.0-beta.3
`astro sync` no longer starts the Cloudflare runtime during type generation, meaningfully reducing sync time for Cloudflare-adapter projects. No configuration changes required.
-
Route caching stabilized in Astro 7 beta 7.0.0-beta.6
Route caching (previously behind `experimental.cache` and `experimental.routeRules`) is now stable and promoted to top-level `cache` and `routeRules` config options. Users on Astro 7 beta must move these keys out of the `experimental` block.
-
Workers: Temporary accounts for AI agent deployments 4.102.0
AI agents can now deploy Workers to Cloudflare without user sign-up using `wrangler deploy --temporary`, creating a 60-minute preview account that can be claimed later. Supports Workers, Static Assets, KV, D1, Hyperdrive, and Queues.
-
Unified Routes page across Mesh, Tunnel, WAN, Magic Transit
The Cloudflare dashboard now consolidates Mesh, Tunnel, WAN, and Magic Transit routes into a single Routes page with an interactive map, unified table, and route testing. Existing APIs and configurations are unchanged; this is a dashboard-only improvement.
-
Codex 0.142.0-alpha.4 pre-release 0.142.0-alpha.4
Fourth alpha pre-release of the upcoming 0.142.0 Codex CLI, indicating active development of the next feature batch. No detailed changelog available yet; monitor for stable release notes.
-
Scheduled tasks/webhooks can no longer approve pending actions 2.1.183
Scheduled task and webhook trigger deliveries were previously treated as keyboard input, meaning they could approve a pending permission prompt or set the session title in auto mode. They are now classified as task notifications and cannot approve actions.
-
MCP auth-stub tools no longer exposed in headless/SDK mode 2.1.183
MCP servers requiring authentication were exposing auth-stub tools to the model in headless/SDK mode, potentially leaking tool signatures or enabling unintended tool calls. This is now fixed.
-
Warning when pinned model is deprecated or auto-updated 2.1.183
Claude Code now shows a warning on stderr in print mode (-p) when the requested model is deprecated or automatically updated to a newer model, including models set in agent frontmatter. Helps fleet operators detect silent model substitutions.
-
Auto mode blocks destructive git/infra commands 2.1.183
Auto mode now blocks destructive git commands (reset --hard, clean -fd, stash drop) when not explicitly requested, blocks amending commits not made by the agent this session, and blocks terraform/pulumi/cdk destroy unless the specific stack was requested. Reduces risk of accidental data loss in automated workflows.
-
Cloudflare IdP is now default for new Zero Trust orgs
New Zero Trust organizations now default to Cloudflare's own identity provider instead of one-time PIN (OTP), restricting auth to existing Cloudflare account members. Existing organizations are unaffected.
-
Cloudflare Fonts input validation and error handling fix
Cloudflare Fonts now adds additional input validation to enhance security and forwards invalid /cf-fonts requests to the origin instead of returning error responses directly. Reduces potential attack surface on malformed font proxy paths.
-
TLS connections now support P-521 certificate signatures rust-v0.141.0
TLS connections now support P-521 certificate signatures commonly used by enterprise proxies, fixing connectivity issues for users behind such proxies. This is relevant for enterprise deployments routing Codex through MITM/inspection proxies.
-
TUI input prompts auto-resolve after inactivity with countdown rust-v0.141.0
TUI input prompts can now auto-resolve after a period of inactivity, displaying a countdown that pauses on user interaction. This enables more automated/unattended Codex workflows.
-
App-server clients can list child threads and redeem rate-limit credits rust-v0.141.0
App-server clients can now list immediate child threads, correlate external-agent imports with detailed results, and read or redeem rate-limit reset credits. This improves multi-agent orchestration and rate-limit management for API consumers.
-
Executor plugins can activate stdio MCP servers per thread rust-v0.141.0
Selected executor plugins can now activate their stdio MCP servers on a per-thread basis, and plugin discovery adds a created-by-me marketplace with auth-specific curated catalogs. This expands the MCP ecosystem integration within Codex.
-
Cloudflare One: Cloudflare IdP now default for new accounts
New Zero Trust organizations now default to the Cloudflare identity provider (account-scoped credentials) instead of one-time PIN (OTP). Existing organizations are unaffected, but teams standing up new Zero Trust orgs should plan for this changed default.
-
getContainerRenderer() root import deprecated in integrations
Importing `getContainerRenderer()` from the package root (e.g. `@astrojs/react`) is now deprecated and logs a warning; it should be imported from the new `@astrojs/react/container-renderer` (and equivalent) sub-path. Affects React, Preact, Svelte, SolidJS, Vue, and MDX integrations.
-
Access: Cloudflare IdP now default for new Zero Trust orgs
New Zero Trust organizations now default to the Cloudflare identity provider instead of one-time PIN (OTP), restricting authentication to Cloudflare account members. Existing organizations are unaffected.
-
Cloudflare Fonts: error handling and security improvements
Cloudflare Fonts now forwards invalid `/cf-fonts` requests to the origin instead of returning 4xx/5xx errors directly, and adds additional input validation to harden against unexpected inputs.
-
Cloudflare Fonts: error handling and security improvements
Cloudflare Fonts now forwards invalid `/cf-fonts` requests to origin instead of returning 4xx/5xx errors directly, and adds additional input validation to enhance security.
-
Hyperdrive: PlanetScale databases billed via Cloudflare
PlanetScale Postgres and MySQL databases can now be created from the Cloudflare dashboard and billed through a Cloudflare account via Hyperdrive. Contract customers gain access in July 2026.
-
PlanetScale DBs creatable & billed via Cloudflare account
You can now create PlanetScale Postgres and MySQL databases from the Cloudflare dashboard and have usage billed through your Cloudflare account via Hyperdrive. This simplifies vendor consolidation for Workers apps using relational databases.
-
Executor plugins can activate stdio MCP servers per thread 0.141.0
Selected executor plugins can now activate their stdio MCP servers on a per-thread basis, and plugin discovery adds a created-by-me marketplace and auth-specific curated catalogs. This expands extensibility for MCP-based tooling workflows.
-
Remote executors use E2E encrypted Noise relay channels 0.141.0
Remote executors now communicate over authenticated, end-to-end encrypted Noise relay channels, replacing unencrypted transport. This materially improves security for users running remote execution workflows.
-
TLS now supports P-521 certs for enterprise proxies 0.141.0
TLS connections in Codex now support P-521 certificate signatures, which are commonly used by enterprise proxies. This unblocks enterprise users who were unable to connect through their corporate network infrastructure.
-
Hyperdrive: PlanetScale Postgres/MySQL via Cloudflare billing
You can now create PlanetScale Postgres and MySQL databases from the Cloudflare dashboard, billed through your Cloudflare account. Hyperdrive manages connection pooling and query caching, enabling Workers apps to use PlanetScale as a relational backend without changing drivers or ORMs.
-
Cloudflare Fonts error handling and security improvements
Cloudflare Fonts now forwards invalid /cf-fonts requests to the origin instead of returning 4xx/5xx directly, and adds additional input validation to enhance security. Sites using Cloudflare Fonts benefit from improved resilience and reduced attack surface.
-
Cross-platform remote execution preserves native working dirs rust-v0.141.0
Cross-platform remote execution now preserves executor-native working directories and shells, including filesystem permission paths across app-server and exec-server boundaries. This improves reliability of remote Codex sessions on mixed OS environments.
-
App-server clients can read and redeem rate-limit reset credits rust-v0.141.0
App-server clients can now list immediate child threads, correlate external-agent imports with detailed results, and read or redeem rate-limit reset credits. This gives programmatic clients better observability and rate-limit management.
-
Codex Rust v0.142.0-alpha.3 pre-release rust-v0.142.0-alpha.3
Third alpha pre-release of the v0.142.0 Rust Codex series; no detailed release notes provided. Competitive intelligence only — not a fleet dependency.
-
Remote executors use E2E encrypted Noise relay channels rust-v0.141.0
Remote executors now communicate over authenticated, end-to-end encrypted Noise protocol relay channels, replacing unauthenticated connections. This materially improves security for any workflow using remote execution.
-
TLS now supports P-521 certs used by enterprise proxies rust-v0.141.0
TLS connections now support P-521 certificate signatures commonly used by enterprise proxies, resolving connectivity failures in corporate network environments.
-
Executor plugins can activate stdio MCP servers per thread rust-v0.141.0
Selected executor plugins can now activate their stdio MCP servers on a per-thread basis, and plugin discovery adds a created-by-me marketplace and auth-specific curated catalogs. This expands MCP integration options for Codex workflows.
-
Codex 0.142.0-alpha.2 Released rust-v0.142.0-alpha.2
Alpha pre-release of Codex 0.142.0-alpha.2 is now available for early testing. No detailed changelog provided in this release entry.
-
Cloudflare Fonts: Error handling & security improvements
Cloudflare Fonts now forwards invalid /cf-fonts requests to the origin instead of returning 4xx/5xx errors directly, and adds additional input validation for security. This reduces information leakage and hardens the fonts proxy against malformed input.
-
Radar Workers AI Popularity Metric Changed to Inference Count
Cloudflare Radar now measures Workers AI model/task popularity by number of inferences rather than unique accounts, affecting both new and historical data via the `/ai/inference/summary/{dimension}` and `/ai/inference/timeseries_groups/{dimension}` API endpoints. Any dashboards or integrations reading these endpoints will see shifted historical trends.
-
astro@7 beta: compressHTML defaults to 'jsx' mode 7.0.0-beta.5
Astro 7 beta now strips whitespace using JSX rules by default (new default: `'jsx'`), which can change rendered output where whitespace between inline elements was previously meaningful. Users must explicitly set `compressHTML: true` or `compressHTML: false` to restore prior behavior.
-
getContainerRenderer() root import deprecated 7.0.0-beta.5
Importing `getContainerRenderer()` from the package root (e.g. `@astrojs/react`) is now deprecated and logs a warning; users should switch to the new `@astrojs/react/container-renderer` entrypoint. Affects React, Preact, Svelte, SolidJS, Vue, and MDX integrations.
-
Codex 0.141.0: Noise relay, cross-platform remote exec, plugin MCP servers 0.141.0
Remote executors now use authenticated end-to-end encrypted Noise relay channels; cross-platform remote execution preserves native working directories and shells. Plugin executor support for per-thread stdio MCP servers and improved marketplace discovery also ship in this release.
-
Codex 0.141.0: SQLite pinned to WAL-reset corruption fix version 0.141.0
Bundled SQLite is now pinned to a version containing the WAL-reset corruption fix, preventing potential data corruption in session storage. This is a material reliability and integrity fix for any deployment using Codex's local session database.
-
Codex Rust 0.142.0-alpha.1 released 0.142.0-alpha.1
Alpha pre-release of Codex 0.142.0 made available; no detailed changelog content provided in this release. Users tracking the alpha channel should monitor for follow-up alpha releases.
-
App-server clients can list child threads and redeem rate-limit credits 0.141.0
App-server clients gain new APIs to list immediate child threads, correlate external-agent imports with detailed results, and read or redeem rate-limit reset credits. This extends programmatic control over multi-agent Codex sessions.
-
TLS now supports P-521 certs used by enterprise proxies 0.141.0
TLS connections in Codex now support P-521 certificate signatures, which are commonly used by enterprise proxy infrastructure. This unblocks users behind corporate MITM proxies that use P-521 certs.
-
TUI input prompts auto-resolve after inactivity timeout 0.141.0
TUI input prompts now auto-resolve after a period of inactivity with a countdown timer that pauses on user interaction, reducing blocked unattended sessions. This is useful for automated or CI-driven Codex workflows.
-
Executor plugins can activate stdio MCP servers per thread 0.141.0
Selected executor plugins can now activate their stdio MCP servers on a per-thread basis, and plugin discovery adds a created-by-me marketplace and auth-specific curated catalogs. This expands the composability of Codex agent workflows.
-
Bundled SQLite pinned to WAL-reset corruption fix version 0.141.0
The bundled SQLite library is now pinned to a version that includes the WAL-reset corruption fix, preventing potential data corruption in session storage. Users relying on Codex session persistence should upgrade.
-
Remote executors use E2E encrypted Noise relay channels 0.141.0
Remote executor channels are now authenticated and end-to-end encrypted using the Noise protocol, replacing previously unencrypted relay connections. This materially improves security for users of remote execution features.
-
Codex Rust 0.141.0-alpha.7 released 0.141.0-alpha.7
Latest alpha release of the Codex Rust rewrite. No detailed changelog provided in this release entry.
-
Bundled Bun runtime upgraded to 1.4 2.1.181
The bundled Bun runtime used by Claude Code has been upgraded to version 1.4, bringing performance improvements and bug fixes to the execution environment.
-
Prompt caching fixed for custom ANTHROPIC_BASE_URL and Foundry 2.1.181
Prompt caching was not reading correctly when using a custom ANTHROPIC_BASE_URL or Foundry, due to a per-request attestation token changing every turn. This is now fixed, restoring cache hit rates and reducing costs for affected deployments.
-
Prompt caching fix for custom ANTHROPIC_BASE_URL and Foundry 2.1.181
Fixed prompt caching not reading on custom `ANTHROPIC_BASE_URL` and on Foundry due to a per-request attestation token changing every turn. Restores caching cost savings for users on custom gateways or Foundry.
-
/config key=value inline syntax for setting configuration 2.1.181
New `/config key=value` syntax allows setting any configuration value directly from the prompt (e.g. `/config thinking=false`), working in interactive, `-p`, and Remote Control modes. Streamlines per-session config changes.
-
Codex Rust 0.141.0-alpha.6 released 0.141.0-alpha.6
Latest alpha iteration of the 0.141 series for Claude Code / Codex Rust CLI. No detailed changelog provided in the feed beyond the version bump.
-
Codex Rust 0.141.0-alpha.5 released 0.141.0-alpha.5
Fifth alpha of the 0.141 series for the Codex Rust CLI. No detailed changelog provided beyond the version bump.
-
Artifacts: Dashboard management for namespaces/repos/tokens
Cloudflare Artifacts (Git-compatible storage on Cloudflare) can now be managed directly from the dashboard, including creating namespaces, repos, forking, and provisioning scoped read/write tokens. Previously only available via API/CLI during beta.
-
Astro 6.4.8: Harden URL decode limits 6.4.8
Hardens the limits on the number of decodings applied to a URL, further tightening the CVE-2025-66202 fix shipped in 6.4.7. This prevents potential edge-case bypasses of the iterative URL-decoding security fix.
-
Codex Rust 0.141.0-alpha.4 released 0.141.0-alpha.4
Alpha pre-release of the next Codex Rust CLI version. No detailed changelog provided in feed entry.
-
Codex Rust 0.141.0-alpha.3 released 0.141.0-alpha.3
Alpha pre-release of the next Codex Rust CLI version. No detailed changelog provided in feed entry.
-
Agents SDK v0.16.1: browser automation, Codemode, recovery 0.16.1
The Agents SDK adds a durable `browser_execute` tool via Browser Run, a new `createCodemodeRuntime` for resumable code execution with approval gates, and client-provided tools for Think sub-agents. Agents can now recover from deploys, Durable Object evictions, and connection churn more reliably.
-
Workers VPC: TCP connect() over VPC Networks
VPC Network bindings now support the connect() Socket API for raw TCP connections to private destinations (Redis, MQTT, custom binary protocols, etc.) via Cloudflare Tunnel/Mesh/WAN on-ramps, in addition to HTTP via fetch(). Previously only HTTP traffic was supported.
-
Codex Rust v0.141.0-alpha.2 released rust-v0.141.0-alpha.2
Alpha pre-release of Codex Rust v0.141.0-alpha.2. No detailed changelog provided in the feed entry.
-
Codex Rust v0.141.0-alpha.1 released rust-v0.141.0-alpha.1
Alpha pre-release of Codex Rust v0.141.0-alpha.1. No detailed changelog provided in the feed entry.
-
Auto mode evaluates subagent spawns before launch 2.1.178
In auto mode, subagent spawns are now evaluated by the safety classifier before launch, closing a gap where a subagent could request a blocked action without review. Improves policy enforcement in automated pipelines.
-
Tool(param:value) syntax for permission rules 2.1.178
Permission rules now support Tool(param:value) syntax to match on a tool's input parameters (with * wildcard), e.g. Agent(model:opus) to block Opus subagents. Enables more granular access control in fleet configurations.
-
TeamCreate/TeamDelete tools removed; implicit team model 2.1.178
The TeamCreate and TeamDelete tools have been removed. With CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, every session now has one implicit team — spawn teammates directly with the Agent tool's name parameter. Any code or workflows calling TeamCreate/TeamDelete will break.
-
Claude Sonnet 4 and Opus 4 retired (returns error)
Claude Sonnet 4 (`claude-sonnet-4-20250514`) and Claude Opus 4 (`claude-opus-4-20250514`) have been fully retired; all API requests to these model IDs now return an error. Migration to Sonnet 4.6 and Opus 4.8 is required.
-
Auto mode: subagent spawns now evaluated before launch 2.1.178
Auto mode now evaluates subagent spawns through the classifier before launch, closing a gap where a subagent could request a blocked action without review. Improves safety of automated multi-agent workflows.
-
Codex Rust 0.140.0 stable: usage views, session delete, Claude Code import, Bedrock auth 0.140.0
Stable 0.140.0 adds /usage token dashboards, permanent session deletion via `codex delete`, /import for migrating Claude Code setup/chats, managed Amazon Bedrock auth with encrypted local credential storage, and a unified @ mentions menu. The /realtime experimental voice feature was removed.
-
Astro 7.0.0-beta: Sätteri is default Markdown processor 7.0.0-beta.4
Astro 7 makes Sätteri (`@astrojs/markdown-satteri`) the default Markdown processor, replacing the remark/rehype pipeline. `@astrojs/markdown-remark` is no longer installed by default, so any project relying on remark/rehype plugins must explicitly install it and set `markdown.processor: unified()`.
-
@astrojs/node 11 beta: Fix non-ASCII filename serving 11.0.0-beta.2
Fixes an issue in the Node adapter where Astro files with non-ASCII characters in their filenames were not correctly served after a production build. Relevant to any Node-hosted Astro app with internationalized file names.
-
Auto mode now classifies subagent spawns before launch 2.1.178
Improved auto mode now evaluates subagent spawns through the classifier before launch, closing a gap where a subagent could request a blocked action without review.
-
Auto mode evaluates subagent spawns before launch 2.1.178
In auto mode, subagent spawns are now evaluated by the classifier before launch, closing a gap where a subagent could request a blocked action without review. Strengthens policy enforcement in automated workflows.
-
Tool(param:value) syntax for permission rules 2.1.178
Permission rules now support `Tool(param:value)` syntax to match on a tool's input parameters (with `*` wildcard), e.g. `Agent(model:opus)` to block Opus subagents. Enables finer-grained access control in settings.
-
WAF: Ghost CMS SQLi CVE-2026-26980 managed rule
Cloudflare added a managed WAF rule blocking a blind SQL injection vulnerability in Ghost CMS Content API (CVE-2026-26980, versions 3.24.0–6.19.0) and a new generic SQLi obfuscated boolean-logic detection rule. Protects edge-proxied Ghost installations automatically.
-
WAF scheduled: Ivanti Sentry command injection rule
A new managed WAF rule for Ivanti Sentry command injection (CVE-2026-10520) is announced for release on 2026-06-22 in Log mode. Users running Ivanti Sentry behind Cloudflare will be passively protected upon release.
-
Codex Rust v0.140.0 stable: usage views, Claude import, Bedrock auth rust-v0.140.0
Stable release adds /usage views for token activity, /import for migrating Claude Code configs and chats, managed Amazon Bedrock authentication, session deletion, and a unified @ mentions menu. Also removes the experimental /realtime voice feature and improves MCP reliability.
-
Tool(param:value) syntax for fine-grained permission rules 2.1.178
Permission rules now support Tool(param:value) syntax (with * wildcard) to match on a tool's input parameters, e.g. Agent(model:opus) to block Opus subagents. Enables more granular allow/deny policies without blocking an entire tool class.
-
JSON schema added to Wrangler config generated by astro add cloudflare astro@6.4.7
Running `astro add cloudflare` now generates a `wrangler.json` (or `wrangler.toml`) with a JSON schema reference, enabling IDE validation and autocomplete for Cloudflare Worker configuration.
-
Fix massive SSR bundle inflation from prerendered redirects astro@6.4.7
Prerendered redirect targets were incorrectly bundled into the SSR function in hybrid mode, causing severe bundle size inflation. The fix ensures only on-demand routes are included in the server bundle.
-
HMR now triggers for src/middleware/ files astro@6.4.7
Changes to files inside `src/middleware/` no longer require a manual dev-server restart; HMR is now correctly triggered for middleware edits during development.
-
Double URL encoding iterative decode replaces 400 rejection astro@6.4.7
Replaces the blanket 400 rejection of double-encoded URLs (which broke tools like Sanity Studio) with iterative decoding that fully resolves percent-encoding before middleware and route matching. This preserves the CVE-2025-66202 middleware authorization bypass fix while restoring compatibility with legitimately double-encoded paths.
-
Sätteri becomes default Markdown processor in Astro 7 astro@7.0.0-beta.4
Astro 7 switches the default `.md` renderer from remark/rehype to the native Sätteri pipeline; `@astrojs/markdown-remark` is no longer installed by default. Projects relying on remark plugins, rehype plugins, or `markdown.remarkRehype` must explicitly install `@astrojs/markdown-remark` and set `markdown.processor: unified()`.
-
Codex permanent session deletion via /delete rust-v0.140.0
Adds `codex delete`, `/delete` command, and app-server `thread/delete` endpoint for permanent session deletion with confirmation safeguards and subagent cleanup. Useful for managing long-running Claude Code/Codex sessions.
-
Codex /usage daily/weekly/cumulative token views rust-v0.140.0
New `/usage` command surfaces daily, weekly, and cumulative account token activity directly in the TUI. Helps teams monitor token consumption against API budgets.
-
Codex /realtime voice controls removed from TUI rust-v0.140.0
The experimental `/realtime` voice controls and related audio dependencies have been removed from the TUI entirely. Any workflows or scripts relying on `/realtime` will break after upgrading.
-
Codex managed Amazon Bedrock API-key auth rust-v0.140.0
Codex now supports managed Amazon Bedrock API-key authentication with encrypted local storage for CLI and MCP OAuth credentials. Useful if using Bedrock as a backend alongside Anthropic API.
-
Codex /import from Claude Code rust-v0.140.0
Codex now supports `/import` to selectively bring in setup, project configuration, and recent chats from Claude Code. Relevant for teams running both Claude Code and Codex side by side.
-
Claude Sonnet 4 and Opus 4 retired (June 15, 2026)
Claude Sonnet 4 (`claude-sonnet-4-20250514`) and Claude Opus 4 (`claude-opus-4-20250514`) have been retired; all requests to these models now return an error. Users must migrate to Claude Sonnet 4.6 or Claude Opus 4.8 respectively.
-
WAF: Cloudforce One threat intel fields in WAF rules
New `cf.intel.ip.*` fields let you match WAF custom and rate-limiting rules against Cloudforce One threat intelligence (DDoS actors, targeted industries/countries). Requires an active Cloudforce One subscription.
-
Fix prerendered redirects inflating SSR bundle in hybrid mode 6.4.7
Prerendered redirect targets were incorrectly bundled into the SSR function in hybrid mode, causing massive bundle size inflation. This fix removes them from the SSR bundle, reducing cold-start size for Workers and Node deployments.
-
Fix CVE-2025-66202 middleware bypass via double URL encoding 6.4.7
Replaces the blanket 400-rejection of double-encoded URLs with iterative decoding so middleware always sees the fully-decoded path, closing the CVE-2025-66202 authorization-bypass vector while restoring compatibility with tools like Sanity Studio that legitimately produce double-encoded URLs.
-
Fix client:only island CSS leaking to unrelated pages 6.4.7
CSS from client:only islands was leaking to unrelated pages when Rollup bundled non-CSS-importing modules into the same chunk as CSS-importing modules. The fix prevents incorrect CSS injection on pages that don't use the affected island.
-
Fix locale URL helpers ignoring trailingSlash:'never' 6.4.7
getRelativeLocaleUrl, getAbsoluteLocaleUrl, and getAbsoluteLocaleUrlList now correctly strip trailing slashes when trailingSlash: 'never' is configured, fixing incorrect URL generation for i18n sites.
-
availableModels allowlist bypass via env vars fixed 2.1.176
Alias model picks could previously be redirected to a blocked model via ANTHROPIC_DEFAULT_*_MODEL environment variables, bypassing the availableModels allowlist. This is now fixed, along with /fast refusing to toggle when it would switch to a disallowed model.
-
enforceAvailableModels managed setting constrains Default model 2.1.175
New `enforceAvailableModels` managed setting ensures the `availableModels` allowlist also constrains the Default model and prevents user/project settings from widening the org's allowlist. Relevant for enterprise orgs controlling model access.
-
footerLinksRegexes setting for custom footer link badges 2.1.176
New `footerLinksRegexes` setting allows regex-matched link badges in the footer row, configurable via user or managed settings. Useful for enterprise deployments embedding context-sensitive links.
-
footerLinksRegexes setting for regex-matched footer link badges 2.1.176
Added `footerLinksRegexes` setting allowing regex-matched link badges in the footer row, configurable via user or managed settings. Enables custom deep-links and tooling integrations surfaced directly in the UI.
-
Bedrock credential caching improved for awsCredentialExport 2.1.176
Credentials from `awsCredentialExport` are now cached until their actual `Expiration` timestamp instead of a fixed 1-hour window. Reduces unnecessary credential refresh overhead for long Bedrock sessions.
-
enforceAvailableModels managed setting for Default model 2.1.175
New enforceAvailableModels managed setting extends the availableModels allowlist to also constrain the Default model selection; user/project settings can no longer widen a managed allowlist. Orgs using availableModels for compliance should enable this to close the Default-model bypass gap.
-
Durable Objects metrics filterable by object ID/name
The Durable Objects Metrics dashboard tab now supports filtering by individual object ID or name, replacing aggregate-only namespace-level views. This makes it easier to debug specific high-traffic objects or error spikes.
-
enforceAvailableModels managed setting added 2.1.175
New `enforceAvailableModels` managed setting constrains the Default model to the `availableModels` allowlist and prevents user/project settings from widening it. Matters for enterprise fleet deployments that restrict model access.
-
Bedrock GovCloud region inference profile prefix fixed 2.1.174
Fixed Bedrock GovCloud regions (`us-gov-*`) deriving the wrong inference profile prefix (`global` instead of `us-gov`), causing 400 errors on derived model IDs. Matters for any fleet using Bedrock in GovCloud.
-
availableModels alias bypass via ANTHROPIC_DEFAULT_*_MODEL fixed 2.1.176
Fixed `availableModels` enforcement: alias model picks can no longer be redirected to a blocked model via `ANTHROPIC_DEFAULT_*_MODEL` environment variables, and `/fast` now refuses to toggle when it would switch to a model outside the allowlist. Closes a bypass path for org model restrictions.
-
Durable Objects: Filter metrics by object ID or name
The Durable Objects Metrics dashboard tab now supports filtering by individual object ID or name, enabling per-object visibility into requests, errors, and storage usage. Previously only namespace-level aggregates were available.
-
Workers AI: Moonshot Kimi K2.7 Code model available
Kimi K2.7 Code (@cf/moonshotai/kimi-k2.7-code), a 1T-parameter MoE code-optimized model with 262k context, is now available on Workers AI. It improves on K2.6 with +21.8% coding benchmark gains and 30% fewer reasoning tokens, at a slightly higher cached-input price of $0.19/M tokens.
-
AI Gateway User Agent Logging
AI Gateway logs now capture the user agent of each client request, enabling identification of which SDK, library, or application sent traffic. Logs can be filtered by user agent in the dashboard.
-
enforceAvailableModels managed setting added 2.1.175
New `enforceAvailableModels` managed setting makes the `availableModels` allowlist also constrain the Default model, preventing user/project settings from widening a managed list. Matters to fleet operators pinning specific model IDs via managed settings.
-
Background sessions no longer inherit other session provider env 2.1.174
Fixed background sessions inheriting another session's `ANTHROPIC_*` provider env vars (gateway URL, custom headers, model aliases) from the shell that started the background daemon. This could cause requests to be routed to the wrong endpoint or with wrong credentials.
-
Bedrock GovCloud region inference profile prefix fix 2.1.174
Fixed Bedrock GovCloud regions (`us-gov-*`) deriving the wrong inference profile prefix (`global` instead of `us-gov`), causing 400 errors on derived model IDs. Critical fix for any fleet using AWS GovCloud with Bedrock.
-
Code execution tool v20260521 discloses 90s time limit code_execution_20260521
The code execution tool version `code_execution_20260521` now discloses the 90-second per-cell execution time limit in its tool description so Claude can budget long-running cells. No beta header required.
-
Web search/fetch tool response_inclusion parameter web_search_20260318
Web search (`web_search_20260318`) and web fetch (`web_fetch_20260318`) tool versions add a `response_inclusion` parameter to drop consumed result blocks from the API response, reducing payload size in agentic workflows. No beta header required.
-
Browser Run /snapshot endpoint multi-format support
The Browser Run /snapshot endpoint now accepts a `formats` parameter to return screenshot, Markdown, and accessibility tree in a single API call. Particularly useful for AI agent workflows that need token-efficient page representations.
-
Dynamic Workers usage count in dashboard and GraphQL API
Cloudflare now shows the number of Dynamic Workers invoked per billing period in the Workers dashboard and via the GraphQL Analytics API using `distinctDynamicWorkerCount`. Helps teams track billable Dynamic Workers usage.
-
DLP: Custom AI Prompt Topics for Zero Trust
Cloudflare DLP now allows defining custom natural-language topics for AI prompt protection, enabling detection of proprietary concepts not covered by predefined categories. Supports ChatGPT, Google Gemini, Perplexity, and Claude — relevant if using Cloudflare Gateway to inspect AI traffic.
-
Workers: Dynamic Workers usage count in dashboard/GraphQL
Customers can now view billable Dynamic Workers counts from the Workers overview dashboard and via GraphQL using the new `distinctDynamicWorkerCount` field. Useful for monitoring billing exposure if using Dynamic Workers.
-
Codex Rust 0.140.0-alpha.9 released rust-v0.140.0-alpha.9
Latest alpha build of the Rust-based Codex CLI (0.140.0-alpha.9) published. No detailed changelog provided in this release entry.
-
Codex Rust v0.140.0-alpha.8 released rust-v0.140.0-alpha.8
Alpha release of Codex Rust 0.140.0-alpha.8. Minimal changelog detail available; part of the ongoing 0.140 alpha series.
-
Sub-agents can now spawn their own sub-agents (5 levels) 2.1.172
Sub-agents can now recursively spawn their own sub-agents up to 5 levels deep, enabling more complex multi-agent workflows. Previously only the top-level agent could spawn sub-agents.
-
Self-hosted sandbox work endpoint on Claude Platform on AWS
The `GET /v1/environments/{id}/work` endpoint for self-hosted sandboxes is now available on Claude Platform on AWS, authorized by the new `GetEnvironment` IAM action.
-
Bedrock now reads AWS region from ~/.aws config files 2.1.172
Amazon Bedrock integration now reads the AWS region from `~/.aws` config files when `AWS_REGION` isn't set, matching standard AWS SDK precedence. Reduces manual configuration for Bedrock-based fleet deployments.
-
Sub-agents can now spawn their own sub-agents (5 levels deep) 2.1.172
Sub-agents can now recursively spawn their own sub-agents up to 5 levels deep, enabling more complex multi-agent workflows. Relevant to teams building dynamic agent pipelines with Claude Code.
-
availableModels restrictions now apply to subagent model overrides 2.1.172
Fixed availableModels restrictions not being applied to subagent model overrides, the agent dispatch model picker, and the advisor model. Previously a subagent could bypass the org allowlist by specifying a model override directly.
-
Sub-agents can now spawn sub-agents (up to 5 levels deep) 2.1.172
Sub-agents can now recursively spawn their own sub-agents up to 5 levels deep, enabling more complex multi-agent workflows. Teams relying on subagent permission policies should verify that nested spawn chains still respect their configured rules.
-
GET /v1/environments/{id}/work available on AWS
The `GET /v1/environments/{id}/work` endpoint for listing pending work on self-hosted sandboxes is now available on Claude Platform on AWS, gated by the new `GetEnvironment` IAM action.
-
availableModels restrictions now apply to subagent model overrides 2.1.172
Fixed `availableModels` restrictions not being applied to subagent model overrides, the agent dispatch model picker, and the advisor model. Ensures org-level model allowlists are enforced across all agent contexts.
-
Sub-agents can now spawn their own sub-agents (5 levels deep) 2.1.172
Sub-agents can now recursively spawn their own sub-agents up to 5 levels deep, enabling more complex multi-agent workflows. Relevant for teams building orchestrated agent pipelines with Claude Code.
-
Bedrock now reads AWS region from ~/.aws config files 2.1.172
Amazon Bedrock now reads the AWS region from `~/.aws` config files when `AWS_REGION` isn't set, matching AWS SDK precedence. Reduces manual config for Bedrock-based fleet deployments.
-
availableModels restrictions now applied to subagent model overrides 2.1.172
Fixed `availableModels` restrictions not being applied to subagent model overrides, the agent dispatch model picker, and the advisor model. Previously, policy could be bypassed via subagent model selection.
-
Sub-agents can now spawn their own sub-agents (5 levels) 2.1.172
Sub-agents can now recursively spawn sub-agents up to 5 levels deep, enabling more complex autonomous workflows. This significantly expands the capability of multi-agent orchestration patterns.
-
Bedrock reads AWS region from ~/.aws config files 2.1.172
Amazon Bedrock now reads the AWS region from `~/.aws` config files when `AWS_REGION` isn't set, matching AWS SDK precedence. Reduces configuration friction for fleet deployments using standard AWS credential files.
-
Self-hosted sandbox work endpoint on Claude Platform AWS
The `GET /v1/environments/{id}/work` endpoint for self-hosted sandboxes is now available on Claude Platform on AWS, authorized via the new `GetEnvironment` IAM action.
-
Cloudflare Images: hosted binding for Workers
The Images binding now exposes `env.IMAGES.hosted` with upload, list, retrieve, update, and delete operations directly from Workers, removing the need for API tokens or HTTP requests. Valuable for any Worker handling image storage.
-
Sub-agents can now spawn sub-agents up to 5 levels deep 2.1.172
Sub-agents can now recursively spawn their own sub-agents up to 5 levels deep, enabling more complex hierarchical agent workflows. Previously sub-agents could not spawn further sub-agents.
-
model attribute added to lines_of_code OTEL metric 2.1.172
The claude_code.lines_of_code.count OTEL metric now includes a model attribute, allowing fleet operators to break down code generation volume by model in their observability dashboards.
-
Codex Rust v0.140.0-alpha.6 released rust-v0.140.0-alpha.6
Alpha release of Codex Rust 0.140.0-alpha.6. Minimal changelog detail available; part of the ongoing 0.140 alpha series.
-
Codex Rust v0.140.0-alpha.5 released rust-v0.140.0-alpha.5
Alpha release of Codex Rust 0.140.0-alpha.5. Minimal changelog detail available; part of the ongoing 0.140 alpha series.
-
Codex Rust v0.140.0-alpha.7 released rust-v0.140.0-alpha.7
Alpha release of Codex Rust 0.140.0-alpha.7. Minimal changelog detail available; part of the ongoing 0.140 alpha series.
-
Bedrock reads AWS region from ~/.aws config when AWS_REGION unset 2.1.172
Amazon Bedrock integration now reads the AWS region from ~/.aws config files when AWS_REGION isn't set, matching standard AWS SDK precedence. Reduces configuration friction for fleet deployments using AWS profiles.
-
Sub-agents can spawn their own sub-agents (up to 5 levels deep) 2.1.172
Sub-agents can now recursively spawn their own sub-agents up to 5 levels deep, enabling more complex multi-agent workflows without manual orchestration. Expands the automation surface for fleet-level agentic tasks.
-
WebFetch wildcard domain rules now correctly match subdomains 2.1.172
Fixed WebFetch(domain:*.example.com) wildcard domain rules never matching subdomains in allow, deny, and ask positions, and file permission rules with mid-pattern wildcards being rejected at startup. Security-relevant: previously wildcarded deny rules silently failed to block requests.
-
Astro addAttribute drops invalid HTML attribute names 6.4.6
astro@6.4.6 now drops attribute names containing characters invalid per the HTML spec (", ', >, /, =, whitespace) in `addAttribute`. This hardens Astro against potential attribute-injection vectors.
-
Astro dev server hot-reloads renamed image files 6.4.6
Renaming an image file while the Astro dev server is running no longer triggers a build error; Astro now correctly hot-reloads the renamed image. Improves local development stability.
-
Astro validates request origin against allowedDomains 6.4.6
astro@6.4.6 validates the request origin against the `allowedDomains` config before fetching prerendered error pages, falling back to localhost when no match. This prevents potential SSRF-style misuse of the Host header during error-page fetching.
-
AI Search namespace management via Wrangler CLI
Wrangler now supports full CRUD namespace commands for AI Search (`wrangler ai-search namespace create/list/get/update/delete`), plus a `--namespace` flag on instance-level commands. Enables scripted and agent-driven namespace management without the dashboard.
-
Codex Rust v0.140.0-alpha.3 released rust-v0.140.0-alpha.3
Alpha pre-release of Codex Rust 0.140.0 (alpha.3). No detailed changelog provided; part of the ongoing 0.140 alpha series.
-
Codex Rust v0.140.0-alpha.4 released rust-v0.140.0-alpha.4
Alpha pre-release of Codex Rust 0.140.0 (alpha.4). No detailed changelog provided in this entry; tracks ongoing 0.140 alpha series.
-
Cloudflare Images: manage hosted images via Workers binding
The Images binding now exposes `env.IMAGES.hosted` with upload, list, retrieve, update, and delete operations, letting Workers manage Cloudflare Images without API tokens or HTTP requests. Useful for any Worker that needs to handle image storage programmatically.
-
DNS Account-Level Records Quota for Enterprise
Cloudflare now enforces DNS record quotas at the account level for Enterprise accounts (1,000,000 records per account for public and internal zones separately) instead of per-zone limits. Non-Enterprise accounts are unaffected and retain existing per-zone quotas.
-
Claude Fable 5 refusal fallbacks opt-in beta
Claude Fable 5 returns `stop_reason: 'refusal'` when safety classifiers block a request; an opt-in `fallbacks` parameter re-runs refused requests on another model at fallback model rates. Not supported on the Message Batches API.
-
Claude Fable 5 requires 30-day data retention (no ZDR)
Claude Fable 5 mandates 30-day data retention on the Claude API and is not available under zero data retention agreements. Organizations with ZDR cannot use Fable 5 without policy changes.
-
Fixed sessions not saving transcripts in VS Code terminal 2.1.170
Sessions launched from the VS Code integrated terminal or shells inheriting Claude Code environment variables were silently not saving transcripts and not appearing in `--resume`. Now fixed.
-
Managed Agents vaults support env-var credentials
Claude Managed Agents vaults now support environment variable credentials, enabling secure injection of secrets (e.g. for CLIs and SDKs) into agent sandboxes without code changes.
-
Claude Fable 5 requires 30-day data retention
Claude Fable 5 requires 30-day data retention on the Claude API and is not available under zero data retention agreements. Organizations with ZDR contracts cannot use this model without a contract change.
-
Fable 5/Mythos 5 tokenizer produces ~30% more tokens
Claude Fable 5 and Mythos 5 use the tokenizer introduced with Opus 4.7; the same text produces roughly 30% more tokens than pre-Opus-4.7 models, which will increase costs and may breach context-window or output-token limits in existing pipelines.
-
Fable 5/Mythos 5: disabled thinking and prefill return 400
On Claude Fable 5 and Claude Mythos 5, `thinking: {type: "disabled"}`, manual extended thinking budgets, and assistant prefill are not supported and return a 400 error. Any fleet code migrating to these models must remove those parameters.
-
New refusal category: reasoning_extraction on Fable 5
The `stop_details.category` field now includes `"reasoning_extraction"` on Claude Fable 5, returned when a request is blocked for attempting to reverse-engineer or duplicate model outputs. No beta header required.
-
Session transcripts now save when launched from VS Code terminal 2.1.170
Fixed sessions not saving transcripts (and not appearing in `--resume`) when launched from the VS Code integrated terminal or any shell that inherited Claude Code environment variables. Ensures audit trails are preserved in IDE-based workflows.
-
Claude Fable 5 (Mythos-class model) now available 2.1.170
Claude Fable 5, a new Mythos-class model, is now generally available in Claude Code. Anthropic states its capabilities exceed any previously generally available model, requiring update to 2.1.170 for access.
-
Fable 5/Mythos 5: adaptive thinking only, no disable claude-fable-5 / claude-mythos-5
On Claude Fable 5 and Mythos 5, `thinking: {type: 'disabled'}` is not supported and returns a 400 error; manual extended thinking budgets and assistant prefill are also unsupported. Code that disables thinking or uses manual budgets must be updated before adopting these models.
-
Managed Agents vaults: environment variable credentials
Claude Managed Agents vaults now support environment variable credentials, enabling secure injection of secrets (CLIs, SDKs, services) into agent sandboxes without exposing them in configuration.
-
Fable 5: 'reasoning_extraction' refusal category added claude-fable-5
The `stop_details.category` field on refusal responses now includes a new `'reasoning_extraction'` value on Claude Fable 5, returned when a request is blocked for attempting to reverse-engineer or duplicate model outputs. No beta header required.
-
WAF: Scheduled rules for Ghost CMS SQLi and obfuscated boolean SQLi (2026-06-15)
Two new WAF rules are scheduled to deploy on 2026-06-15 in Log mode: Ghost CMS SQLi (CVE-2026-26980) and a generic obfuscated boolean SQLi rule targeting URI inputs. Teams running Ghost CMS should be aware of the upcoming log-mode rule.
-
WAF: New rules for Drupal SQLi, Axios prototype pollution, Mirasvit PHP injection
Cloudflare WAF Managed Ruleset gained new blocking rules for CVE-2026-9082 (Drupal/PostgreSQL SQLi), CVE-2026-40175 (Axios prototype pollution), and CVE-2026-45247 (Mirasvit Cache Warmer PHP object injection). The Axios CVE is directly relevant to Node.js-based services in the fleet.
-
Codex 0.139.0: Code mode standalone web search rust-v0.139.0
Code mode can now call standalone web search directly, including from nested JavaScript tool calls, receiving plaintext results. Tool/connector input schemas now also preserve oneOf/allOf for better MCP tool compatibility.
-
Codex 0.139.0: Sandbox proxy-only networking enforced rust-v0.139.0
Sandbox execution now preserves approved escalation decisions and more consistently enforces configured proxy-only networking, closing a gap where sandbox could bypass network restrictions. Relevant to any fleet use of Codex with sandboxed execution.
-
Codex Rust v0.139.0 stable: web search in code mode rust-v0.139.0
Code mode can now call standalone web search directly (including from nested JS tool calls). Also fixes sandbox proxy-only networking enforcement and improves MCP tool schema compatibility with oneOf/allOf.
-
Claude Fable 5 (Mythos-class model) available in Claude Code 2.1.170
Claude Fable 5 is a new Mythos-class model described as exceeding capabilities of any previously generally available model. Accessible by updating to version 2.1.170 or later.
-
WAF: Axios prototype pollution CVE-2026-40175 blocked
Cloudflare Managed Ruleset now blocks CVE-2026-40175, a prototype pollution vulnerability in the Axios HTTP client library that can cause denial-of-service or code execution. Also adds coverage for Drupal PostgreSQL SQLi (CVE-2026-9082) and Mirasvit Cache Warmer PHP object injection (CVE-2026-45247).
-
Codex Rust v0.140.0-alpha.2 released rust-v0.140.0-alpha.2
Alpha pre-release of Codex Rust v0.140.0-alpha.2. Minimal release notes; signals active development on the next minor version.
-
Fix: sessions not saving transcripts in VS Code terminal 2.1.170
Fixed sessions launched from the VS Code integrated terminal (or any shell inheriting Claude Code env vars) not saving transcripts and not appearing in --resume. Directly affects VS Code-based fleet workflows.
-
Claude Managed Agents scheduled deployments (cron)
Claude Managed Agents now supports scheduled deployments via cron, allowing sessions to run on a schedule without operators managing their own scheduler infrastructure.
-
Fable 5 requires 30-day data retention; no ZDR claude-fable-5
Claude Fable 5 mandates 30-day data retention on the Claude API and is unavailable under zero data retention (ZDR). Operators with ZDR agreements cannot use this model without a policy change.
-
Fable 5/Mythos 5: thinking:disabled & prefill removed claude-fable-5
On Claude Fable 5 and Mythos 5, `thinking: {type: 'disabled'}`, manual extended thinking budgets, and assistant prefill all return a 400 error. Code relying on these patterns will break when targeting these models.
-
Fable 5 refusal: new reasoning_extraction category claude-fable-5
The `stop_details.category` field now includes `'reasoning_extraction'` on Claude Fable 5, returned when a request is blocked for attempting to reverse-engineer model outputs. No beta header required; existing refusal handlers should be updated to handle the new category.
-
Fable 5 tokenizer: ~30% more tokens vs pre-Opus-4.7 claude-fable-5
Claude Fable 5 uses the tokenizer introduced with Opus 4.7, which produces approximately 30% more tokens than pre-Opus-4.7 models for the same text. This affects cost estimates, context window utilization, and any hardcoded token budgets.
-
Codex Rust v0.140.0-alpha.1 released rust-v0.140.0-alpha.1
Alpha release of Codex Rust v0.140.0-alpha.1. No detailed changelog content was provided in this entry.
-
Codex v0.139.0: Sandbox proxy-only networking fix rust-v0.139.0
Sandbox execution now preserves approved escalation decisions and enforces configured proxy-only networking more consistently, closing a gap where network policy could be bypassed.
-
Codex v0.139.0: Code mode standalone web search rust-v0.139.0
Code mode can now call standalone web search directly, including from nested JavaScript tool calls, returning plaintext results. Tool/connector input schemas now preserve oneOf and allOf for better MCP tool compatibility.
-
Codex Rust CLI v0.139.0 stable release rust-v0.139.0
Stable release of Codex Rust CLI v0.139.0. Follows the alpha series (alpha.1–alpha.3) published the same day.
-
Fix sessions not saving transcripts from VS Code terminal 2.1.170
Sessions launched from the VS Code integrated terminal (or any shell inheriting Claude Code env vars) were not saving transcripts and not appearing in --resume. This is now fixed.
-
Claude Fable 5 (Mythos-class model) now available 2.1.170
Claude Fable 5 is a new Mythos-class model exceeding all previously generally available models in capability. Available via update to 2.1.170, with a linked announcement post.
-
Claude Fable 5 and Claude Mythos 5 Launch claude-fable-5 / claude-mythos-5
Claude Fable 5 (widely available) and Claude Mythos 5 (Project Glasswing) launch with 1M token context, 128k max output, always-on adaptive thinking, and a new tokenizer producing ~30% more tokens than pre-Opus-4.7 models. Breaking changes include no support for `thinking: {type: 'disabled'}`, manual thinking budgets, or assistant prefill, and a 30-day data retention requirement.
-
Claude Fable 5 New Tokenizer (~30% More Tokens) claude-fable-5
Claude Fable 5 uses the tokenizer introduced with Claude Opus 4.7; the same text produces roughly 30% more tokens than pre-Opus-4.7 models, directly affecting prompt cost, context window usage, and any token-budget logic in fleet code.
-
Claude Fable 5 Refusal Stop Reason & No-Billing
Claude Fable 5 returns `stop_reason: 'refusal'` when safety classifiers decline a request, and adds a new `stop_details.category` value `'reasoning_extraction'` for ToS violations around model output reverse-engineering. Requests refused before any output is generated are not billed.
-
WAF: New rules for CVE-2026-9082, CVE-2026-45247, CVE-2026-40175
Cloudflare WAF Managed Ruleset adds blocking rules for a Drupal/PostgreSQL SQL injection (CVE-2026-9082), Mirasvit Cache Warmer PHP object injection (CVE-2026-45247), and Axios prototype pollution (CVE-2026-40175). Also adds disabled generic SQLi obfuscated-boolean detection rules.
-
WAF: Scheduled Ghost CMS SQLi & obfuscated boolean rules (2026-06-15)
Two new WAF rules scheduled to deploy on 2026-06-15 in Log mode: a Ghost CMS SQL injection detection (CVE-2026-26980) and a generic SQLi obfuscated boolean URI rule. These rules will be live shortly after announcement.
-
@astrojs/node experimental.logger startup message fix 10.1.4
Fixes the `experimental.logger` destination not being used for the server startup message in @astrojs/node. The logger is now resolved before the server starts listening, ensuring the correct destination is used.
-
Sätteri processor gains Prism syntax highlighting 0.3.0
@astrojs/markdown-satteri now supports Prism syntax highlighting by setting `markdown.syntaxHighlight: 'prism'`, removing the previous Shiki-only restriction. This expands the usability of the Rust-based Sätteri markdown processor.
-
@astrojs/markdown-satteri Rust-based processor launched 0.2.0
A new optional Markdown processor package `@astrojs/markdown-satteri` based on the Sätteri Rust pipeline is now available, offering significantly faster builds than the default Remark-based processor. Astro plans to make it the default processor in the future.
-
Cloudflare adapter sets immutable cache headers for static assets 13.7.0
The @astrojs/cloudflare adapter now automatically injects immutable Cache-Control headers for static assets under the _astro directory at build time, improving CDN cache performance without any user configuration.
-
@astrojs/db integration deprecated 0.21.3
The @astrojs/db integration is now deprecated as the team no longer has bandwidth to maintain it. Users are recommended to migrate directly to a database client of their choice (Drizzle, Kysely, etc.).