Repository navigation
meta(changelog): Update changelog for 11.13.0 - #26426
Conversation
**Why:** Removing rimraf in #26219 left two unused lockfile entries that pnpm prunes during installation, causing develop's lockfile stability check to fail. Commit the generated cleanup to restore CI. Co-authored-by: OpenAI Codex <codex@openai.com>
…ecords the visitor IP (#26177) Fixes #26154 Server tunnels sent the visitor IP to Sentry only as `X-Sentry-Forwarded-For`. Sentry SaaS replaces that header with the connecting IP, so events sent through a tunnel stored the server's IP. The tunnel now also passes through the incoming `CF-Connecting-IP`, `X-Vercel-Forwarded-For` and `X-Forwarded-For` headers, which SaaS keeps and prefers over it. A rewrite proxy sends the same headers. `X-Sentry-Forwarded-For` stays for self-hosted Relay. Tested against sentry.io (de region) with a Next.js handler mode app. Before the fix, a request behind Cloudflare and one with only `X-Forwarded-For` both stored the server IP. After the fix, both store the client IP. Behind a proxy that appends to `X-Forwarded-For`, a client can pick the first entry and so its stored IP. A rewrite proxy behaves the same, and events can already set `user.ip_address`. Not tested on Cloudflare Workers or Vercel Edge, which may change these headers on outgoing requests. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…ock streams (#26192) ## What Bedrock ConverseStream and InvokeModelWithResponseStream spans now collect stream data into the shared stream state and end through `endStreamSpan`, like the other AI integrations. Stream spans now also get `gen_ai.response.streaming` and a total token count for all model families. ## Why The shared state is the base for recording Bedrock outputs in a later PR, and it makes Bedrock stream spans consistent with the other AI integrations. Extracted from #26118. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…tarts (#26224) **Why:** Nitro imports the server entry lazily, in the first request. Without `node --import`, an `import './instrument.server'` in `server.ts` runs `Sentry.init()` only after that request started, so every cold start loses its `http.server` span. This follows what `@sentry/nuxt` does since #24094: the Vite plugin finds `instrument.server.{ts,mts,js,mjs}` (project root, then `src/`), bundles it into the Nitro build and loads it from a Nitro plugin, which runs when the server starts. Like Nuxt, this is always on and has no option. Orchestrion already instruments the drivers at build time, so `--import` is no longer needed in production. In `vite dev` you still need it. Existing setups keep working. A `--import` preload or an import in `server.ts` now runs `init()` a second time. The first client wins, and the SDK logs that the preload or import can be removed (Nuxt logs the same). `init()` is also skipped while TanStack Start prerenders (`TSS_PRERENDERING`), which Nuxt does too. Otherwise the bundled config would send events from the build. The `tanstackstart-react` e2e app now starts without `--import`. The whole suite passes locally that way, including the ioredis and mysql spans. `tanstackstart-react-static` keeps `--import`, so it covers existing start commands. **How to review:** Start with `src/vite/serverInstrumentation.ts`, then `init()` in `src/server/sdk.ts`. **Unsure about:** The docs still tell people to copy `instrument.server.mjs` and preload it, or to import it in `server.ts`. They need a follow-up in sentry-docs, like the Nuxt migration note. Closes JS-4051 <!-- junior-request-attribution:start --> via **Francesco Gringl-Novy**. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:slack%3AD0CA8FXFXAS%3A1794297657.485969 --> -- [View Session](https://junior-prod.sentry.dev/conversations/slack%3AD0CA8FXFXAS%3A1794297657.485969) [[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AD0CA8FXFXAS%3A1794297657.485969/?project=4510944073809921) <!-- junior-session-footer:end --> --------- Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Francesco Gringl-Novy <francesco.novy@sentry.io> Co-authored-by: Charly Gomez <charly.gomez1310@gmail.com>
**Why:** The JSDoc for `enableLongAnimationFrame` says `Default: false`, but `DEFAULT_BROWSER_TRACING_OPTIONS` sets it to `true`. This fixes only the comment and has no runtime change. It was split out from #26141, which was closed in favor of #26253. <!-- junior-request-attribution:start --> via @BYK. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:agent%3Aaf6d95d56ac3e36093e1be6d251b295c --> -- [View Session in Sentry](https://sentry.sentry.io/explore/conversations/agent%3Aaf6d95d56ac3e36093e1be6d251b295c/?project=4510944073809921) <!-- junior-session-footer:end --> Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: BYK <126780+BYK@users.noreply.github.com>
[Gitflow] Merge master into develop
…aps (#26092) **Why:** The new `sentry` CLI throws a `ValidationError` for any directory that has no JS + source map pairs. `SentryCliAdapter.uploadSourcemaps` uploads directories one at a time, so the first empty directory stops the whole loop. `@sentry/nextjs` uploads `.next/server` first, so if server source maps are off, the client chunks that do have maps never get uploaded. The plugin only logs the error and the build passes, so this goes unnoticed. CLI 0.48+ stops reporting this error to Sentry, but it still throws, so the upload still stops without this change. The CLI check stays strict on purpose, because it catches build misconfiguration. The adapter now passes `allowEmpty: true`, collects the directories that uploaded 0 files, and logs a warning for each. In direct-upload mode, "Successfully uploaded" is logged only if at least one directory actually uploaded something. The staged-upload path already returns early with a warning when no artifact has a source map, so it is unchanged. Before: ``` [@sentry/nextjs - Node.js] Error: An error occurred. Couldn't finish all operations: ValidationError: Found 214 JS file(s) in '.next/server' but no companion .map files. [...] Pass --allow-empty to suppress. # .next/static/chunks is never uploaded ``` After: ``` [@sentry/nextjs - Node.js] Warning: No source maps found in ".next/server", so nothing was uploaded from it. This usually means source map generation is not enabled for this build output. Enable it so Sentry can un-minify stack traces from these files. [@sentry/nextjs - Node.js] Info: Successfully uploaded source maps to Sentry # .next/static/chunks is uploaded ``` **How to review:** Start with `src/core/cli.ts`, then the two call sites in `build-plugin-manager.ts`. `cli-contract.test.ts` runs the real CLI (now `sentry@^0.49.0` after #26200) against the mock server with one empty directory followed by one real directory. **Unsure about:** A directory that does not exist still throws and stops the loop; `allowEmpty` only covers existing directories without source maps. I left that case out of this PR. <!-- junior-request-attribution:start --> via **bete**, @betegon. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:local%3Aweb%3A667ad986d499815a40e93973 --> -- [View Session in Sentry](https://sentry.sentry.io/explore/conversations/local%3Aweb%3A667ad986d499815a40e93973/?project=4510944073809921) <!-- junior-session-footer:end --> --------- Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Miguel Betegon <miguel.betegon@sentry.io> Co-authored-by: bete <28983162+betegon@users.noreply.github.com>
…nner streams on 1.2+ (#25931) On `@anthropic-ai/sdk` 1.2.0+, streamed turns from `beta.messages.toolRunner({ stream: true })` and `beta.messages.thread().stream()` had no `gen_ai.chat` spans. Every tool runner runs over a thread since 1.2.0, including on Bedrock and Vertex. The thread calls `beta.messages.create()`, which opens our span, but parses the response through a new `APIPromise` our hook never sees, so the span never ended. This hooks `BetaMessageStream._start()`, matches each request to its stream by abort signal, and ends the span from the stream's events with usage and output. `instrumentMessageStream` now also ends the span on `end`, so aborted or truncated streams, `messages.stream()` included, no longer leave it open. An abort before the response is not marked as an error. Fixes #25816 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ter upload in the esbuild plugin (#25648) Part of the stack for #25609. esbuild writes its output before `onEnd` runs and has no hook to drop an output, so the plugin sets `sourcemap: 'external'` when the setting is unset and deletes the maps after a successful upload: every map in the metafile when the plugin turned them on, and exactly the maps the upload read from disk when the user enabled them, which `uploadSourcemaps` now reports. When the upload fails all maps stay, with an error log. With `write: false` the setting is left alone, because nothing reaches the disk. The metafile lists outputs relative to `absWorkingDir`, while the upload globbed them from the process working directory, so builds with an `absWorkingDir` elsewhere uploaded nothing. The paths are now resolved against `absWorkingDir` first, and passed with forward slashes because glob reads backslashes as escapes. Integration fixtures cover `ignore` and `assets` on the staged path for esbuild.
…dflare Pages (#26188) This PR addresses a few issues I found while investigating some comments on #25999 ### CF Pages runs the browser SDK on the server This means it will immediately crash if we ship anything that assumes a server-side SDK is being used, this showed up when I tried to use the `serveTunnelIntegration` on the other PR, it doesn't exist on the client SDKs. This happens because we do not have an `exports` condition for `worker/workerd` like other SDKs. So `@sentry/astro` resolves to `index.client.js`. ```js // @sentry/astro exports["."]: no workerd entry, so workerd falls through to the client build { "node": "./build/esm/index.server.js", "browser": "./build/esm/index.client.js", "import": "./build/esm/index.client.js" } // @astrojs/cloudflare v12 SSR resolve conditions: no "node" vite.ssr.resolve.conditions.push("workerd", "worker"); ``` We had no E2E apps covering this so it never came up, also it doesn't happen while using `withSentry`. But a user trying to call `Sentry.init` from `@sentry/astro` will trigger the bug. I fixed this by adding a new `workerd/worker` export conditions that point to a new `index.worker.ts`, which rexports the cloudflare SDK and uses the `init` from there. There is a different way to fix it via vite plugins, but I preferred the more "standard" way here. --- ### Cloudflare Client Events are dropped if `withSentry` is not used This showed up while testing the first issue, the tests were timing out. This is because in CF pages, the `withSentry` wrapper is never used so our sends were never registered with `waitUntil`, so the `flush` returned early, cutting off the requests. I fixed this by making the `flush` wait for the previous sends until they are drained. Which also fixes the issue for every possible usage of `@sentry/cloudflare` that isn't wrapped by `withSentry`. ```ts // _registerWithInvocationWaitUntil const pending = Promise.resolve(promise).then(() => undefined, () => undefined); this._pendingEagerDrains.add(pending); void pending.then(() => this._pendingEagerDrains.delete(pending)); // flush(), after super.flush(timeout) await Promise.all(this._pendingEagerDrains); ``` There is no risk of hanging here because each send is already capped by 2 seconds. --- ### The Workers wrapper can shadow the `sentry.server.config` client Because of the first fix, we now have an order problem. A `withSentry` wrapper can override the client that was loaded from `sentry.server.config` which would break all events, since there is no DSN. The override happens because the wrapper runs first. It's not really an override, more of a "shadowing". This didn't happen before because the first client loaded from the browser bundle and so they never conflicted. The fix is to ensure the order loads the "correct" init first, which can be arbitrary. I did the following to de-risk the idea: - If no `sentry.server.config` file exists, then it is never added at the top of the import order. So the wrapper client is the only source. - If both `sentry.server.config` file exists along with a wrapper client, prefer the server file client. - The Astro worker `init` reads the bindings from `cloudflare:workers` and fills unset options the same way the wrap does, so an init without a `dsn` still picks up `SENTRY_DSN`. This only lives in the `workerd` entry since `cloudflare:workers` doesn't resolve elsewhere. I did think about applying (3) to the cloudflare SDK in general but that may break apps that run it under Node, so need to figure it out later.
OAuth registration can identify a different application from MCP `clientInfo`, but transport `AuthInfo` has no standard client name. Add an opt-in `getOAuthClientName` resolver that records the registered name as `mcp.auth.client.name`, independently of `mcp.client.name`. For example, when a registered application named `Example Desktop` connects through a client identifying as `example-connector`, before the span contains only `mcp.client.name = "example-connector"`. After configuring the resolver, it also contains `mcp.auth.client.name = "Example Desktop"`. Resolve the name once per incoming request, including discovery and unsupported methods. The callback receives `clientId` and application-defined `extra` metadata when available and may read request-scoped authentication context. Only the returned name is recorded; resolver failures leave dispatch intact. Preserve the wrapper's existing first-explicit-value option behavior. Real HTTP coverage exercises both MCP SDK majors in ESM/CJS, concurrent requests, and missing identities. #25059 is merged; this branch is rebased on develop and imports `MCP_AUTH_CLIENT_NAME` from published `@sentry/conventions@0.27.0`. The conventions publication blocker is resolved. This identity distinction follows [MCP client registration](https://modelcontextprotocol.io/specification/2026-07-28/basic/authorization/client-registration). Network metadata (#24506) and canonical attributes (#25060) remain separate changes. Validated the 48-project SDK build, formatting and lint, 151 MCP unit tests, and ten real MCP SDK v1/v2 integration tests after rebase.
…el shim (#26239) **Why:** With the TanStack Start tunnel enabled, the server entry is loaded through a stand-in module that only does `export * from "<entry>?sentry-tunnel-original"`. The Cloudflare auto-instrument plugin treats both modules as the entry. The classes get wrapped in the real one, but the stand-in has no named classes, so it printed a false "Could not auto-instrument" warning. The plugin now skips the warning when a module re-exports everything from its own file under a query. Closes #26204 Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
**Why:** The org-wide Warden workflow (`getsentry/.github`) only runs `security-review` here. We have ~25 workflows, several of which drive AI agents (`auto-fix-issue`, `dependabot-auto-triage`, `track-framework-updates`, `optional-e2e`), plus ~20 agent skills managed via `agents.toml`. This adds a repo `warden.toml` that layers two more skills on top of the org base: `gha-security-review` (from `getsentry/skills`) for `.github/workflows/**` and `.github/actions/**`, and the already-vendored `skill-scanner` for `.agents/**`, `.claude/**`, `.cursor/**`, and `agents.toml`/`agents.lock`. sentry-dotnet and sentry-python run the same skills. `failOn = "high"` matches most getsentry repos; with `failCheck` and `requestChanges` left at their `false` defaults, findings are still reported only and never block merges. Each skill is scoped with explicit `paths` because repo-layer skills don't inherit the org `ignorePaths`. `model` is left unset so it inherits the org default; the org workflow only provides an OpenRouter key. **How to review:** Single config file. I validated it with `@sentry/warden@0.53.0` (`loadWardenConfig` + `resolveSkillConfigs`), and both skills resolve (local `skill-scanner`, remote `gha-security-review`). It hasn't run end-to-end in CI yet. Since this PR touches neither workflows nor skills, neither new skill should trigger on it. **Unsure about:** Whether `.cursor/**` (BUGBOT.md, environment.json) belongs in skill-scanner scope. I included it because it's agent instructions. <!-- junior-request-attribution:start --> via **Francesco Gringl-Novy**. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:slack%3AD0CA8FXFXAS%3A1794381869.399109 --> -- [View Session](https://junior-prod.sentry.dev/conversations/slack%3AD0CA8FXFXAS%3A1794381869.399109) [[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AD0CA8FXFXAS%3A1794381869.399109/?project=4510944073809921) <!-- junior-session-footer:end --> --------- Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Francesco Gringl-Novy <francesco.novy@sentry.io>
Bumps [node-forge](https://github.com/digitalbazaar/forge) from 1.4.2 to 1.4.3. <details> <summary>Changelog</summary> <p><em>Sourced from <a href="https://github.com/digitalbazaar/forge/blob/main/CHANGELOG.md">node-forge's changelog</a>.</em></p> <blockquote> <h2>1.4.3 - 2026-11-03</h2> <h3>Security</h3> <ul> <li><strong>HIGH</strong>: X.509 certificate chain verification ignores name constraints. <ul> <li><code>pki.verifyCertificateChain()</code> does not check the nameConstraints extension (RFC 5280 4.2.1.10). A non-critical nameConstraints extension in a CA certificate is ignored, so a name-constrained intermediate CA can issue certificates for names outside its permitted subtrees, or inside its excluded subtrees, and the chain verifies. Exploitation requires the private key of a name-constrained CA that chains to a trusted root. A critical nameConstraints extension is rejected as an unsupported critical extension.</li> <li>Reported by <a href="https://github.com/bhaswanthc"><code>@bhaswanthc</code></a>.</li> <li>CVE ID: <a href="">CVE-TBD</a></li> <li>GHSA ID: <a href="https://github.com/digitalbazaar/forge/security/advisories/GHSA-3cc4-3294-vgjf">GHSA-3cc4-3294-vgjf</a></li> </ul> </li> <li><strong>MODERATE</strong>: TLS client crash on a handshake record without a session. <ul> <li>A TLS client throws an uncaught <code>TypeError</code> from <code>process()</code> when it receives a ServerHello while it has no session: after a <code>close_notify</code> alert or a call to <code>close()</code>, or before <code>handshake()</code> is called. A malicious server, or an on-path attacker, can terminate a Node.js process that passes socket data to <code>process()</code> without catching exceptions. Servers are not affected.</li> <li>Reported by <a href="https://github.com/manus-use"><code>@manus-use</code></a>.</li> <li>CVE ID: <a href="">CVE-TBD</a></li> <li>GHSA ID: <a href="https://github.com/digitalbazaar/forge/security/advisories/GHSA-j8rw-jp99-435v">GHSA-j8rw-jp99-435v</a></li> </ul> </li> <li><strong>MODERATE</strong>: ReDoS in <code>forge.http</code> response header parsing. <ul> <li><code>forge.http</code> trims response header and <code>Set-Cookie</code> values with a regular expression that takes quadratic time on a long run of whitespace followed by another character. A server, or an on-path attacker, can block a <code>forge.http</code> client for minutes with one large header.</li> <li>Reported by <a href="https://github.com/manus-pi"><code>@manus-pi</code></a>.</li> <li>CVE ID: <a href="">CVE-TBD</a></li> <li>GHSA ID: <a href="https://github.com/digitalbazaar/forge/security/advisories/GHSA-6cv4-p8m7-4r25">GHSA-6cv4-p8m7-4r25</a></li> </ul> </li> <li><strong>MODERATE</strong>: TLS client crash on an unsupported ServerHello compression method. <ul> <li>A TLS client accepts any compression method in the ServerHello, although it only offers no compression. An unknown method makes <code>process()</code> throw an uncaught <code>Error</code> at ServerHelloDone, before the server is authenticated. A malicious server, or an on-path attacker, can terminate a Node.js process that passes socket data to <code>process()</code> without catching exceptions, including through <code>forge.tls.wrapSocket</code>. Servers are not affected.</li> <li>Reported by <a href="https://github.com/alienkeric"><code>@alienkeric</code></a>.</li> <li>CVE ID: <a href="">CVE-TBD</a></li> <li>GHSA ID: <a href="https://github.com/digitalbazaar/forge/security/advisories/GHSA-w4xr-hqv8-ch44">GHSA-w4xr-hqv8-ch44</a></li> </ul> </li> <li><strong>LOW</strong>: Ed25519 signature verification accepts small-order public keys. <ul> <li><code>ed25519.verify()</code> does not reject public keys of small order, such as the identity point. Under such a key, a signature with <code>S = 0</code> and a small-order <code>R</code> verifies for many or, with the identity key, all messages, without any private key. Applications are affected only if</li> </ul> </li> </ul> <!-- raw HTML omitted --> </blockquote> <p>... (truncated)</p> </details> <details> <summary>Commits</summary> <ul> <li><a href="https://github.com/digitalbazaar/forge/commit/936b24d939dc7261ee37982576f9cf2b410084ee"><code>936b24d</code></a> Release 1.4.3.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/4d8f79afe9661aedf1c5a1a28463d9564905df91"><code>4d8f79a</code></a> Update changelog.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/1d9816d333c23c523ef2bd8b24282c87db2654ef"><code>1d9816d</code></a> Check the ServerHello compression method.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/326a4d0903ee46358bc849a5c3fd96d9c21489bd"><code>326a4d0</code></a> Trim forge.http header values in linear time.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/2cfaeb516cb3d4aff1c0569d39394df41c5cb17c"><code>2cfaeb5</code></a> Ignore handshake messages on a TLS client without a session.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/6b23eae8beac316b6867bdb0e7ad7c2c959a97f6"><code>6b23eae</code></a> Complete nameConstraints enforcement.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/862aa5f630bdf40b0dfda408d6ac357bf35a7d6b"><code>862aa5f</code></a> Enforce nameConstraints in <code>verifyCertificateChain()</code>.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/956f2b511709cdf37d16180354a0ae7e1d947725"><code>956f2b5</code></a> Document ed25519.verify() small-order key behavior.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/1b77e6f20082d7484479303d0252d5f427651295"><code>1b77e6f</code></a> Reject small-order Ed25519 public keys in verify.</li> <li><a href="https://github.com/digitalbazaar/forge/commit/2bd5146c042aabcba86977372befbd80f89c418e"><code>2bd5146</code></a> Add GitHub usernames to credits.</li> <li>Additional commits viewable in <a href="https://github.com/digitalbazaar/forge/compare/v1.4.2...v1.4.3">compare view</a></li> </ul> </details> <br /> [](https://docs.github.com/en/github/managing-security-vulnerabilities/about-dependabot-security-updates#about-compatibility-scores) Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting `@dependabot rebase`. [//]: # (dependabot-automerge-start) [//]: # (dependabot-automerge-end) --- <details> <summary>Dependabot commands and options</summary> <br /> You can trigger Dependabot actions by commenting on this PR: - `@dependabot rebase` will rebase this PR - `@dependabot recreate` will recreate this PR, overwriting any edits that have been made to it - `@dependabot show <dependency name> ignore conditions` will show all of the ignore conditions of the specified dependency - `@dependabot ignore this major version` will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself) - `@dependabot ignore this minor version` will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself) - `@dependabot ignore this dependency` will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself) You can disable automated security fix PRs for this repo from the [Security Alerts page](https://github.com/getsentry/sentry-javascript/network/alerts). </details> --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com>
…ckfile (#26270) npm deprecated `node-abort-controller@3.1.1` yesterday, which broke `pnpm dedupe --check` on develop because the lockfile stores deprecation notices, fixed by adding the notice to the lockfile.
**Why:** CODEOWNERS plus team auto-assignment requests every matching team, so a PR that touches several areas pings a lot of people. This workflow instead tops each PR up to exactly 3 human reviewers when it is opened, reopened, or marked ready for review. It picks people at random from the matching codeowner teams first, then fills from `@getsentry/team-javascript-sdks`. It never picks the author, `mydea`, or `stephanie-anderson`. Manually added reviewers count toward the 3. Bot reviewers don't. Once a PR has 3 reviewers, the workflow doesn't change anything. The workflow reads `.github/CODEOWNERS` as it is. **Before merging:** on the browser, server and framework sub-teams, turn off "Enable auto assignment" and turn on "Only notify requested team members". If auto assignment stays on, GitHub adds its own picks first, the workflow counts them, and it stops. GitHub will still add the code owner team requests to each PR. The notify setting keeps those from pinging the whole team. Depends on: * getsentry/security-as-code#3904 * getsentry/security-as-code#3903 **How to review:** Start with `scripts/assign-reviewers.js`. The workflow is a thin `github-script` wrapper. **Unsure about:** - The token comes from the existing `GITFLOW_APP_*` app because listing team members needs org `members: read`, which `GITHUB_TOKEN` can't get. Check that the app has `members: read` and `pull_requests: write`, or swap in another app. - If none of the 3 picks belongs to a matching codeowner team, that team may still be notified as a whole. This can happen when a PR touches 2 areas. - The `team-javascript-sdks` ruleset (required reviewers, 1 approval, `*`) may also request the whole team on its own. - The workflow uses `pull_request_target` so it also runs on fork PRs. It only checks out the base branch, so PR code never runs with the token. <!-- junior-request-attribution:start --> via **Francesco Gringl-Novy**. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:slack%3AD0CA8FXFXAS%3A1794382310.154169 --> -- [View Session](https://junior-prod.sentry.dev/conversations/slack%3AD0CA8FXFXAS%3A1794382310.154169) [[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AD0CA8FXFXAS%3A1794382310.154169/?project=4510944073809921) <!-- junior-session-footer:end --> --------- Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Francesco Gringl-Novy <francesco.novy@sentry.io>
**Why:** If neither `url` nor `SENTRY_URL` is set, the plugin passes `https://sentry.io` to the CLI. A self-hosted `sntrys_` org token records the server it came from, so the CLI refuses to send it to sentry.io. The build still passes, but nothing gets released or uploaded. With `url` left unset, the CLI uses the token's server for `sntrys_` tokens and sentry.io otherwise, so SaaS setups behave the same. **Before / After:** self-hosted `sntrys_` token created on `sentry.example.com`, with no `url` in the plugin config. Before: ``` [sentry-vite-plugin] Error: An error occurred. Couldn't finish all operations: HostScopeError: Credentials: https://sentry.io Refusing to route requests here because it doesn't match the host your Sentry credentials are for (https://sentry.example.com). To use this host, run: sentry auth login --url https://sentry.io To keep using your current credentials, remove this URL override. ``` After: ``` # no error: the release and source maps go to https://sentry.example.com ``` **How to review:** The fix is one line in `options-mapping.ts`. The new `cli-contract.test.ts` case runs the real CLI with a `sntrys_` token whose server is the mock server and no `url`. It fails with the error above before this change. **Unsure about:** `NormalizedOptions.url` is now `string | undefined`. It's internal, and only `cli.ts` and `telemetry.ts` read it. The telemetry check still treats an unset `url` as SaaS, which is today's behavior. <!-- junior-request-attribution:start --> via @betegon. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:agent%3A497276480faf8cef7b74e6b05edf0932 --> -- [View Session in Sentry](https://sentry.sentry.io/explore/conversations/agent%3A497276480faf8cef7b74e6b05edf0932/?project=4510944073809921) <!-- junior-session-footer:end --> --------- Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: bete <28983162+betegon@users.noreply.github.com> Co-authored-by: Miguel Betegón <miguelbetegongarcia@gmail.com>
…bort (#26236) **Why:** When a `messages.stream()` call is aborted, the Anthropic SDK's `MessageStream` emits only `abort` and then `end`. It never emits `message` or `error`. We only ended the span on `message` or `error`, so an aborted stream's span never ended and was never sent. The SDK always emits `end` last, so we now end the span there too. `endStreamSpan` does nothing if the span already ended. The span keeps whatever usage and response data streamed in before the abort. Inside `end`, an aborted stream (`stream.aborted`) is now marked `cancelled`, the same as Google GenAI when the caller's `abortSignal` fires. Only the caller's `stream.abort()` or request `signal` sets `aborted`. The SDK's request timeouts are only active until the response headers arrive, and they surface as `APIConnectionTimeoutError` on `error`, so they still record `internal_error`. We deliberately don't listen for `abort`. The SDK raises an unhandled rejection for an abort only when nothing listens for it, so an `abort` listener would silence that rejection for apps that abort a stream and never await it. `stream.aborted` exists in every supported SDK version (`>=0.19.2`). **How to review:** The `end` listener in `instrumentMessageStream` is the whole change. I also checked it by hand against the real `@anthropic-ai/sdk` `MessageStream`, aborting it mid-stream against a local SSE server without awaiting it. Uninstrumented and with this change, the app gets the same single unhandled rejection. With this change, the span also ends `cancelled` with the partial usage attached. **Note:** Under span streaming (the default), `getSimpleStatus` sends `cancelled` as `ok` and drops the status message, as it does for every cancelled span in the SDK. Static mode shows `cancelled`. An aborted span still differs from a completed one: it has no finish reason, and if it was aborted before the response, it has no response ID or usage. <!-- junior-request-attribution:start --> via **Rola Abuhasna**. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:slack%3AD0C848BQ509%3A1794305469.823259 --> -- [View Session](https://junior-prod.sentry.dev/conversations/slack%3AD0C848BQ509%3A1794305469.823259) [[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AD0C848BQ509%3A1794305469.823259/?project=4510944073809921) <!-- junior-session-footer:end --> --------- Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Rola Abuhasna <rola.abuhasna@sentry.io>
…crumbs (#26201) Why: `fetch('')` is instrumented, but `xhr.open('GET', '')` was not, and once an empty URL request is instrumented its breadcrumb shows `url: ""`. This PR fixes both * `xhr.ts` bailed on `!url`, which treats an empty string like an unparseable argument, so empty-URL XHRs got no breadcrumb and no span. It now only bails on `url === undefined`, so XHR matches fetch. * The fetch and XHR breadcrumb handlers now show the page URL for an empty URL instead of `""`. Query and fragment are stripped with `stripUrlQueryAndFragment`, so params like `?token=` from the page URL don't end up in breadcrumbs. Since an empty URL resolves to the page URL itself, the helper uses `location.href` directly instead of calling `new URL()` (bcoz its throws for hrefs like `about:srcdoc`). without a `location` object (e.g. workers), nothing changes. Only empty URLs are rewritten, _absolute_ and _relative_ URLs keep their current breadcrumb behaviour. How to review: Start with: * `packages/browser-utils/src/instrumentation/xhr.ts` * `packages/browser/src/integrations/breadcrumbs.ts` and the regression coverage in: * `packages/browser-utils/test/instrumentation/xhr.test.ts` * `packages/browser/test/integrations/breadcrumbs.test.ts` The key things to check are that only empty URLs are rewritten, and that the query and fragment are stripped from the page URL in all three breadcrumb paths (fetch, failed fetch, XHR). Unsure about: Whether relative URLs (like `/api/foo?x=1`) should also be resolved against `location.href` and stripped in breadcrumbs. I kept this limited to empty URLs so the change stays scoped, but yeah I'm happy to widen it. Ref [#13168](#13168)
This PR adds the external contributor to the CHANGELOG.md file, so that they are credited for their contribution. See #26201 Co-authored-by: s1gr1d <32902192+s1gr1d@users.noreply.github.com>
…26233) **Why:** Function names can be arbitrary strings. Emscripten, for example, names its bridge functions `methodCaller< (emscripten::val) => void>`. The Chrome parser treated the first ` (` as the start of the location, which garbled the filename and broke sourcemap resolution. The Firefox/Safari parser has an optional `(...)` group that strips legacy argument lists (`bar(1)@…`). That group could stretch across several parenthesised parts of a name and truncate it. This is the browser counterpart of #26216. **How to review:** Both fixes are regex tweaks in `stack-parsers.ts`. Chrome: a negative lookahead skips a `(` when a `) ` follows before the next `(` opens. Gecko: the argument group can no longer contain parentheses. The existing wasm test had locked in the broken output, so its expectations changed. Closes #26205 --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
`withSentry` and `WorkerEntrypoint` instrumentation covered `tail`, but not the other two tail handlers that workerd still supports. workerd calls `trace` only when a worker does not export `tail`. Streaming tail workers export `tailStream`. Errors thrown by either handler were not captured. `trace` now gets the same wrapper as `tail`. `tailStream` needs more: workerd calls it once with the onset event, then calls the function or object it returns for each later batch of events. The wrapper also wraps that return value, so each event handler call runs as its own invocation and captures its errors. The mechanism types are `auto.faas.cloudflare.trace` and `auto.faas.cloudflare.tail_stream`, to match the existing `auto.faas.cloudflare.tail`. Closes #24038 Closes JS-3573 --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…er (#26264) In `vite dev`, the id of a module that is not pre-bundled ends with `?v=<hash>`. `@apm-js-collab/code-transformer-bundler-plugins` 0.7.5 matched module paths exactly, so the orchestrion transform skipped packages that the user excluded from `optimizeDeps`. Version 0.7.6 removes the query before it matches. The bump adds 0.7.6 to `minimumReleaseAgeExclude`, because the release is only a few days old. A new unit test transforms an excluded package with a `?v=<hash>` id. It fails on 0.7.5 and passes on 0.7.6. The `cloudflare-vercelai-v7` e2e app now also runs its test in `vite dev`. The app uses Vite 7, so this covers the esbuild dep optimizer plugin, and it shows that `ai` sends `gen_ai.*` spans in dev. The test fails when `transformOptimizedDeps` is off. `vite dev` gets the DSN only with `CLOUDFLARE_INCLUDE_PROCESS_ENV=true`. Closes #25765 Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…#26265) Before the framework SDKs move from their own `filesToDeleteAfterUpload` globs to the plugin's `deleteSourcemapsAfterUpload`, the E2E tests should catch any source map left in the output. This adds the checks that were missing. - `nextjs-sourcemaps` gets two variants that use the SDK default (delete after upload), one with Turbopack and one with webpack. Both check that `.next/static` has no `.map` files and no `sourceMappingURL` comments. The base app keeps `deleteSourcemapsAfterUpload: false`. - `nextjs-app-dir` and `nextjs-pages-dir` read only the top level of `.next/static`, where there are no files, only folders, so their map check could never fail. They now search the whole folder. - `tanstackstart-react-sourcemaps` also checks `.output/public` for `sourceMappingURL` comments. - Gatsby has no E2E app. A unit case pins the current behaviour for `devtool: false`, which the SDK replaces with `hidden-source-map` and then deletes.
**Why:** `node-flue (latest)` fails before any test runs:
`@flue/runtime` 2.3.0 (latest) does `import { Type } from
"@earendil-works/pi-ai"`, its `^1.1.0` range now resolves to pi 1.5.x,
and pi 1.5.0 dropped the `Type` re-export
([withastro/flue#912](withastro/flue#912),
still open). The pinned variant (Flue 2.0.5 / pi 0.83) is not affected.
Same root cause as #26235 for `cloudflare-pi-durable`.
The overrides use `@^1` selectors, so they only match the latest
variant. I checked locally with pnpm 12.11.2: the pinned install still
resolves pi 0.83.0, latest resolves pi-ai/pi-agent-core 1.4.0, and `vite
build` passes again. Drop the override once Flue ships a fix.
**Unsure about:** Pinning pi-agent-core too, so pi-ai doesn't end up
duplicated (pi-agent-core 1.5.1 pins pi-ai 1.5.1 exactly).
<!-- junior-request-attribution:start -->
via **Francesco Gringl-Novy**.
<!-- junior-request-attribution:end -->
<!-- junior-session-footer:start -->
<!-- junior-conversation-id:slack%3AD0CA8FXFXAS%3A1794311940.741409 -->
--
[View
Session](https://junior-prod.sentry.dev/conversations/slack%3AD0CA8FXFXAS%3A1794311940.741409)
[[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AD0CA8FXFXAS%3A1794311940.741409/?project=4510944073809921)
<!-- junior-session-footer:end -->
---------
Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com>
Co-authored-by: Francesco Gringl-Novy <francesco.novy@sentry.io>
…undles (#26277) **Why:** The `solidstart-2` e2e test has been failing on `develop` since #26200 bumped the Sentry CLI to `^0.49.0`. The server entry of `@sentry/solidstart` still re-exports the deprecated SolidStart 1 `withSentry`. That pulls `@sentry/bundler-plugins`, and with it the CLI and `oxc-parser`, into the Nitro server graph. Starting with 0.48, the CLI ships as code-split ESM. Rolldown still emits its dynamic-import chunks after tree-shaking the code that imports them; Rollup doesn't. Nitro also traces `oxc-parser`'s native binding. For every SolidStart 2 app, the result is about 3.6 MB of unused build tooling in `.output/server` (5.3 MB → 1.7 MB with this change). The CLI chunk also carries the CLI's own debug IDs, which the app can't upload, and that's the assertion that fails. Removing the root `withSentry` export is breaking (#25915, v12), and lazy imports are off the table. So `sentrySolidStart()` now adds a `sentry-solidstart-runtime` export condition to server environments. The exports map resolves that condition to the existing server entry, which is the same as today's minus `withSentry`. SolidStart 1's `app.config.ts` never sets the condition, so it still gets `withSentry`. **How to review:** Start with the `exports` change in `package.json`, then `makeServerEntryPlugin`. Locally, `solidstart-2` (including the tunnel-basepath dev variant) and `solidstart` pass against CI tarballs. **Unsure about:** Whether a custom export condition is the shape we want here, versus living with the bloat until v12. The plugin only extends conditions that are already set, because Vite would otherwise replace its defaults. Nitro and SolidStart always set them for their server environments. Separately, the Rolldown behaviour seems worth an upstream issue. <!-- junior-request-attribution:start --> via **Francesco Gringl-Novy**. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:slack%3AD0CA8FXFXAS%3A1794301479.516639 --> -- [View Session](https://junior-prod.sentry.dev/conversations/slack%3AD0CA8FXFXAS%3A1794301479.516639) [[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AD0CA8FXFXAS%3A1794301479.516639/?project=4510944073809921) <!-- junior-session-footer:end --> Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Francesco Gringl-Novy <francesco.novy@sentry.io>
**Why:** Allow newly published `@sentry/*` packages and the unscoped `sentry` CLI immediately without adding a version-specific release-age exemption for each update. Remove the expired OpenTelemetry API exemption; other packages retain their existing release-age policy. --------- Co-authored-by: OpenAI Codex <codex@openai.com>
**Why:** When an x64 dependency cache is missing, its key can prefix-match the ARM64 cache, restoring incompatible native bindings and breaking tests before they start. Include architecture in every cache key so x64 and ARM64 caches cannot match each other. Co-authored-by: OpenAI Codex <codex@openai.com>
…6287) **Why:** The runner images don't include the Node.js version pinned in `devEngines.runtime` (24.22.1). As a result, almost every CI job has `actions/setup-node` fetch its version manifest from `actions/node-versions` through the GitHub API, which costs 2 REST calls per job and counts against the repo's `GITHUB_TOKEN` limit. A full run has about 370 jobs, so at busy hours we run out. setup-node falls back to nodejs.org on its own, but steps without a fallback then fail with `API rate limit exceeded for installation`: Get CI Metadata (which fails the whole run), setup-bun, the optional E2E action, Size Check, Warden and Dependency Review. Restoring the install from the Actions cache lets setup-node find it locally, so it makes no API calls. On a cache miss, it behaves like today. **How to review:** Start with `.github/actions/setup-node/action.yml`. Everything else just swaps `actions/setup-node` with `node-version-file: 'package.json'` for the new action. Matrix jobs (`matrix.node`, `'24'`) and the E2E app's own Node setup are unchanged. **Unsure about:** Only `develop` saves the cache, so all PRs share one entry per architecture (about 47 MB each), because the repo cache is already at about 10.75 GB. PRs won't hit the cache until this has run once on `develop`. On that first run, parallel jobs race to save, so all but one log a non-fatal "unable to reserve cache" warning. caching may be slower than resolving this every time, but this is something we'll need to keep an eye on and evaluate when it runs. <!-- junior-request-attribution:start --> via **Francesco Gringl-Novy**. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:slack%3AD0CA8FXFXAS%3A1794381869.399109 --> -- [View Session](https://junior-prod.sentry.dev/conversations/slack%3AD0CA8FXFXAS%3A1794381869.399109) [[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AD0CA8FXFXAS%3A1794381869.399109/?project=4510944073809921) <!-- junior-session-footer:end --> Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Francesco Gringl-Novy <francesco.novy@sentry.io>
This PR adds the external contributor to the CHANGELOG.md file, so that they are credited for their contribution. See #26382 Co-authored-by: andreiborza <168741329+andreiborza@users.noreply.github.com> Co-authored-by: Martin Sonnberger <martin.sonnberger@sentry.io>
## What The Bedrock `awsIntegration` now records the model response in `gen_ai.output.messages` for Converse, ConverseStream, InvokeModel, and InvokeModelWithResponseStream. This includes text, reasoning, and tool calls, for all model families. ## Why Bedrock spans showed the prompt but not the response. Extracted from #26118. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
… tokens (#25834) Gemini reports the tokens of built-in tool results fed back to the model (Google Search, URL context, code execution) in `usageMetadata.toolUsePromptTokenCount`, apart from `promptTokenCount`. `totalTokenCount` includes both ("Total token count for prompt, response candidates, and tool-use prompts (if present)" in `types.ts`). `addResponseAttributes` and the stream's `handleResponseMetadata` only used `promptTokenCount` for `gen_ai.usage.input_tokens`, so with those tools `input_tokens + output_tokens` did not add up to `gen_ai.usage.total_tokens`. The tool-use prompt tokens are now added to the input count, for a response and for a stream (the two counts are tracked separately, so a chunk that repeats only one of them keeps the other). Responses without tool-use tokens are unchanged. Tests in `google-genai-streaming.test.ts` cover a response and a stream with `toolUsePromptTokenCount` (input 140, total 160) and a response without it. The two with tool-use tokens fail on `develop`. This change was made with Claude Code. --------- Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com> Co-authored-by: Andrei Borza <andrei.borza@sentry.io>
This PR adds the external contributor to the CHANGELOG.md file, so that they are credited for their contribution. See #25834 Co-authored-by: andreiborza <168741329+andreiborza@users.noreply.github.com>
…26325) Next.js is about to send the trace in a `Server-Timing` header (#26107). This pr just prepares the reusable functions for the consuming SDKs (Next, Remix, Nitro) . `isSharedCacheable` takes a header reader so it works on a fetch `Response` and on a Node `ServerResponse` alike. Refs JS-3864 --------- Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
The `Server-Timing` value and the shared-cache check now live in core (previous PR of the stack). Remix drops its own copies and calls the core helpers. Behavior is unchanged; the middleware tests cover the cache headers as before. Refs JS-3864 Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
…t responses (#26107) Stacked on the core and Remix PRs; this one holds the Next.js change only. **Why:** Under `cacheComponents` the `sentry-trace` meta tag arrives with the resumed part of the document, usually after the browser SDK started, so the pageload has to wait for it (#25142). A `Server-Timing` header reaches the browser before any HTML. The browser SDK already reads it, and Remix and Nitro already send it. On a self-hosted server Next.js flushes the headers with the first chunk, inside the request span, so the header is set in `writeHead` rather than when the request comes in. Before, with the meta tag only: ```mermaid sequenceDiagram participant B as Browser participant S as Next.js server B->>S: GET /page Note over S: request span starts, shell comes from cache S-->>B: headers + shell (no trace context) Note over B: SDK starts, pageload has to wait Note over S: dynamic part renders S-->>B: resumed part with sentry-trace meta tag Note over B: pageload starts late, fetches and errors before it are unattached ``` After, with the header: ```mermaid sequenceDiagram participant B as Browser participant S as Next.js server B->>S: GET /page Note over S: request span starts, shell comes from cache Note over S: writeHead on the first chunk: Server-Timing carries the active trace S-->>B: headers + shell (Server-Timing: sentry-trace, baggage) Note over B: SDK starts, pageload continues the trace at once Note over S: dynamic part renders S-->>B: resumed part with sentry-trace meta tag Note over B: tag only settles page.server_rendering.type ``` The header is left off responses a shared cache may store, which is what Next.js sends for static and ISR app router documents. A CDN in front of the server would otherwise replay one visitor's trace and sampling decision to everyone. The check is the one the Remix 3 middleware already had, now shared as `isSharedCacheable` in core. Non-document responses get no header either. Since the pageload no longer waits for the tag, `page.server_rendering.type` is settled after span start for a runtime shell with a dynamic part, so `tracesSampler` does not see it in that case. The observer stops when the pageload span ends, so a dynamic part that outlasts the pageload leaves the type unset. Closes #24945 Refs JS-3864 --------- Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
…Integration (#26414) ## What `awsIntegration({ recordInputs, recordOutputs })` controls whether Bedrock prompts and responses are recorded. The options take precedence over `dataCollection.genAI` and apply only to Bedrock Runtime calls. ## Why The other AI integrations take these options, so users can turn off Bedrock content recording without turning it off for every AI integration. Extracted from #26118. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…26419) ## What Adds an Important Changes entry: Bedrock spans now record prompts and responses by default, and `awsIntegration` takes `recordInputs` and `recordOutputs`. ## Why Users who send Bedrock calls will see new content on their spans after they upgrade. The entry tells them how to turn it off. Closes: #26118 Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
## What Adds astro-7 E2E tests for how server islands connect to the pageload trace today: GET islands on static and SSR pages, and a POST island. ## Why Baseline for server island support. A GET island's `<link rel="preload">` leaves before the SDK runs, so its render starts its own trace, while a POST island joins the pageload trace. Linear: [JSSDK-38](https://linear.app/getsentry/issue/JSSDK-38/astro-server-islands) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…ndex (#25835) The Workers AI streaming instrumentation keyed every tool-call delta by `toolCall.index ?? 0`. A stream that omits `index` (or sends `null`) and sends each parallel call in its own chunk was recorded as one call, with the first call's id and name and every call's arguments concatenated. The other calls were missing from `gen_ai.response.tool_calls` and `gen_ai.output.messages`. #25622 fixed the same bug for the OpenAI integration. Workers AI itself seems to always send `index`, and Cloudflare's own `workers-ai-provider` also keys by `index ?? 0`. So this guards OpenAI-compatible routes rather than fixing a captured Workers AI stream. The key selection from #25622 now lives in `ai/core/streaming-tool-calls.ts`, shared by both integrations. It also matches a known `id` before grouping by `index`, so a call that starts without an index and continues with one stays one call, in OpenAI too. **Known limits.** No model is known to stream these shapes, so there is no heuristic for them: - With no `index` and no `id`, a name repeated on every fragment, or an `id` that only arrives on a later fragment, splits one call into two. Develop gave one call here only because it merged everything into index 0. - A fragment with a numeric `index` but no `id`, following a call opened without an index, starts a new call. - The same `id` on two different indexes merges them into one call. No linked issue, same class of bug as #25622. This change was prepared with the assistance of an AI coding agent (Claude). --------- Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com> Co-authored-by: Rola Abuhasna <rola.abuhasna@sentry.io>
…26397) ## What Next.js now writes a `sentry-render-start` meta tag (whole UNIX seconds) next to `sentry-cache-origin`. The browser uses it to set `page.server_rendering.age` on the pageload span, so the UI can show how old the cached HTML was. The browser side is in `@sentry/core`, so SvelteKit and Nuxt only need to write the tag. They follow in stacked PRs. ## Why The `cache_origin` link tells you which render produced a cached page, but not how old that page is. That linked span is often not sampled or already gone, so the browser needs the start time itself. The tag holds the start of the linked render span. Without a recording span, it holds the current time, which is the render start, because Next.js computes the meta tags at the start of `prerenderToStream`. The attribute comes from getsentry/sentry-conventions#799, which is not merged yet, so the name is a local constant with a TODO. A `Server-Timing` header (#26107) does not replace the tag: shared-cacheable responses and Vercel do not send it, and it cannot hold the render time of a cached shell.
#26398) Stacked on #26397. ## What SvelteKit now writes the `sentry-render-start` meta tag next to `sentry-cache-origin` on prerendered and Vercel ISR pages. The browser sets `page.server_rendering.age` from it, as in #26397. ## Why The tag holds the start of the linked render span. Without a recording span (build prerender, unsampled request), it holds the request start that `sentryHandle` records. The current time would be too late by the whole render, because Kit calls `transformPageChunk` only after the render. `_INTERNAL_getCacheOriginMetaTags` is the new shared writer for both tags. The old `_INTERNAL_getCacheOriginMetaTag` stays until Nuxt moves to the new one in the next PR.
… example (#26408) ## What The documented server entry now forwards TanStack Start's request options to the handler. The e2e fixtures use the same shape, and a new test checks that request context set in the server entry reaches a server route. ## Why Our example called `handler.fetch(request)` and dropped the second argument. TanStack Start reads the typed request context from it, so apps that copied our setup lost that context. Part of #26022. Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
`zodErrorsIntegration` was written against Zod 3 and misses two Zod 4 shapes. `zod/mini` throws an error named `$ZodError`, so those errors skipped the integration entirely. Zod 4 also nests union failures under `errors` instead of `unionErrors`, so they ended up in `extra` as deeply nested objects (and got truncated by normalization) instead of a flattened string. This change accepts `$ZodError` and stringifies `errors` the same way we already stringify `unionErrors`. The `errors` key is only added when an issue actually has it, so Zod 3 output stays byte-for-byte the same. The new tests import `zod/v4` and `zod/v4/mini`. Those subpaths exist in both the current 3.25 dev dependency and in 4.x, so the tests don't depend on #25655 landing first. <!-- junior-request-attribution:start --> via **Andrei Borza**. <!-- junior-request-attribution:end --> <!-- junior-session-footer:start --> <!-- junior-conversation-id:slack%3AC0BKK9QLCUA%3A1793095128.386479 --> -- [View Junior Session](https://junior-prod.sentry.dev/conversations/slack%3AC0BKK9QLCUA%3A1793095128.386479) [[Sentry]](https://sentry.sentry.io/explore/conversations/slack%3AC0BKK9QLCUA%3A1793095128.386479/?project=4510944073809921) <!-- junior-session-footer:end --> Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com> Co-authored-by: Andrei Borza <andrei.borza@sentry.io> Co-authored-by: Andrei <168741329+andreiborza@users.noreply.github.com>
…6399) Stacked on #26398. ## What Nuxt now writes the `sentry-render-start` meta tag next to `sentry-cache-origin` on cached and prerendered pages. The browser sets `page.server_rendering.age` from it, as in #26397. ## Why The tag holds the start of the linked render span. Without a recording span, it holds the request start. Nitro calls `render:html` only after the whole render, so a Nitro `request` hook records the start on `event.context`. The hook also fires for prerender requests, and a background SWR revalidation sees the context of the request that started it. On Cloudflare, this also gives the right start when the async context is lost.
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Want fixes drafted automatically? Bugbot Autofix can create code changes for findings. A team admin can enable Autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 600cf87. Configure here.
| uses: oven-sh/setup-bun@v2 | ||
| with: | ||
| bun-version: '1.3.14' | ||
| bun-version: ${{ inputs.bun-version || '1.4.2' }} |
There was a problem hiding this comment.
Conflicting Bun version defaults
Medium Severity
The bun-version input defaults to 1.3.14, but the setup step falls back to 1.4.2 when that input is empty. E2E jobs always pass matrix.bun-version, which is empty for unpinned apps, so those runs install 1.4.2 instead of the oldest supported Bun. deno-version keeps a matching default and fallback.
Reviewed by Cursor Bugbot for commit 600cf87. Configure here.
size-limit report 📦Compared: fd06345 (PR test merge using master at ec12330) Against: 8bebd29 (master) · base run #65564
|
Optional E2E testsSome optional checks were not successful. They do not block the merge. 5 not successful, 58 successful. See the run.
Successful checks
|


No description provided.