Skip to content

chore(repo): version packages (beta) - #1017

Open
github-actions[bot] wants to merge 1 commit into
v3from
changeset-release/v3
Open

github-actions[bot] wants to merge 1 commit into
v3from
changeset-release/v3

Conversation

@github-actions

@github-actions github-actions Bot commented Nov 7, 2026 •

Copy link
Copy Markdown
Contributor

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to v3, this PR will be updated.

⚠️⚠️⚠️⚠️⚠️⚠️

v3 is currently in pre mode so this branch has prereleases rather than normal releases. If you want to exit prereleases, run changeset pre exit on v3.

⚠️⚠️⚠️⚠️⚠️⚠️

Releases

@evlog/eve@1.0.0-beta.0

Major Changes

  • #1134 6e1b1ac Thanks @HugoRCD! - New package: evlog for eve 1.0, mounted as an eve extension with export { default } from '@evlog/eve' in agent/extensions/evlog.ts. Each turn emits one wide event with the id sessionId:turnId, read from the session eve stores, so a turn resumed in another process is complete. Tools add fields with useLogger() from @evlog/eve/logger, and a thrown evlog error keeps its code, status, why and fix on the event. Records are redacted in every environment, and record: 'minimal' | 'standard' | 'debug' with per-tool rules decides what a turn carries.

evlog@3.0.0-beta.3

Major Changes

  • #1111 c3cf3e1 Thanks @HugoRCD! - A second log.error() on the same request or job logger now replaces the first error instead of merging into it. Before, the event took the second error's message and code but kept fields only the first had, such as its fix or cause, so it described neither error. Fields you set on error yourself with log.set(), or through the error() context argument, are kept.

    To keep both failures, record the earlier one under its own key: log.set({ receipt: { error } }).

  • #1002 8d0cbde Thanks @HugoRCD! - On macOS and Linux, the file system drain (evlog/fs) creates its directories with mode 0700 and its log files with mode 0600, so only the account that runs the app can read the events. Before, both followed the process umask, usually 0755 and 0644, which let every account on the machine read them. Windows is unchanged: Node applies only the write bit there, so the files keep the permissions inherited from their folder.

    A directory or file that already exists keeps its mode. Run chmod -R go= .evlog to restrict logs written before the upgrade. Another account can no longer open the new files. A log shipper running as its own user should run as the app's account, or receive the events through an HTTP drain. A container running as root that writes into a bind-mounted project on Linux now leaves files your own account cannot read: run it with --user "$(id -u):$(id -g)".

  • #1131 8219cdf Thanks @HugoRCD! - The AI block on wide events moves from ai to gen_ai, for evlog/ai and evlog/eve. Fields take the OpenTelemetry GenAI semantic convention names where the conventions define one (gen_ai.request.model, gen_ai.provider.name, gen_ai.usage.input_tokens, gen_ai.usage.cache_read.input_tokens, gen_ai.response.finish_reasons), and evlog's own fields sit under the same key in snake_case (gen_ai.calls, gen_ai.usage.cost, gen_ai.usage.estimated_cost, gen_ai.tools[].duration_ms). AILogger.getMetadata() and onUpdate return the same shape. eve turn events also carry gen_ai.conversation.id, gen_ai.agent.name and gen_ai.operation.name.

    Migrate queries, dashboards and alerts: ai.model becomes gen_ai.request.model, ai.inputTokens becomes gen_ai.usage.input_tokens, ai.costUsd becomes gen_ai.usage.cost, ai.estimatedCost becomes gen_ai.usage.estimated_cost, and ai.finishReason becomes the array gen_ai.response.finish_reasons. The full table is on the AI SDK metadata page. The OTLP drain now sends gen_ai.usage.cache_write.input_tokens instead of gen_ai.usage.cache_creation.input_tokens.

  • #1038 fb90d8a Thanks @HugoRCD! - Three optional peer ranges stop at the majors evlog is tested on:

    • @nestjs/common and @nestjs/core go from >=11.2.8 to ^11.2.8 || ^12.0.0.
    • react-router goes from >=7.18.5 to ^7.18.5 || ^8.0.0.
    • hono goes from >=4.13.12 to ^4.13.12 || ^5.0.0-rc.1. evlog/hono runs on Hono 4 and on the Hono 5 release candidates. Hono 5 calls app.onError as middleware, (c, next) with the error on c.error, so update your own error handler when you move to it.

    NestJS 13, React Router 9 and Hono 6 now produce a peer warning until a release of evlog adds them.

  • #1134 6e1b1ac Thanks @HugoRCD! - Removes the evlog/eve entry point. Install @evlog/eve, replace agent/hooks/evlog.ts (defineEvlogHook(config)) with agent/extensions/evlog.ts (export default evlog(config) from @evlog/eve), and import useLogger from @evlog/eve/logger. The migration guide lists the removed options.

    evlog/toolkit adds onErrorCreated(listener), called with every EvlogError as it is built, and createMiddlewareLogger().finish() accepts durationMs for an integration that knows when the work started.

Minor Changes

  • #1036 9663cdd Thanks @HugoRCD! - The Better Stack adapter now sends a message, as Better Stack's own @logtail/node client does. Live tail headlines a line with it and Better Stack builds _pattern from it, so request events now group by route. An event without message gets the headline the pretty console prints, followed by the error message, for example POST /api/checkout 402 in 104ms Payment failed. An event with nothing to headline gets no message. A message already on the event, from a tagged log or log.set({ message }), is sent unchanged.

  • #1020 a3385c7 Thanks @HugoRCD! - evlog/fastify attaches the request logger to request.evlog, typed as ForkableLogger, so request.evlog.set(), .fork() and .audit() type-check. Fastify types request.log as its own logger, and a module augmentation cannot change the type of an existing property: request.log.set() failed with TS2339 even though it worked at runtime, and with skipLibCheck: false the published declarations reported TS2717. request.log still holds the evlog logger at runtime, so existing code keeps working. On a route that include or exclude filters out, or with enabled: false, request.evlog accepts every call and emits nothing. On @nestjs/platform-fastify, evlog/nestjs sets request.evlog the same way.

  • #1110 e2ac261 Thanks @HugoRCD! - Add errorHandler: false to the evlog/nuxt, evlog/nitro and evlog/nitro/v2 modules. evlog's Nitro error handler is then not registered, so your own Nitro errorHandler or the framework's default answers errors. The framework default can include the error's data, which carries why, fix and link, for example in development, so set your own errorHandler when they must stay private. Errors still reach the wide event. The default is unchanged.

  • #1119 854529a Thanks @HugoRCD! - Nitro v3, Nitro v2 and Nuxt: the request logger now has log.fork(label, fn), and useLogger() works without the event.

    log.fork() runs background work under a forked logger, passed to fn, that emits its own wide event with operation and _parentRequestId. The child goes through the evlog:enrich and evlog:drain hooks like the request event, and is registered with waitUntil where the runtime provides one:

    export default defineEventHandler(async (event) => {
      const log = useLogger(event);
      log.fork("send_receipt", async (child) => {
        child.set({ receipt: { sent: true } });
        await sendReceipt();
      });
      return { ok: true };
    });

    With Nitro's request context turned on (experimental: { asyncContext: true } in nitro.config.ts, or nitro.experimental in nuxt.config.ts), useLogger() with no argument returns the current request logger, so services can reach it without the event. Without the flag it throws an error that names it. useLogger(event) is unchanged and works on every runtime.

  • #1067 b8eb0de Thanks @HugoRCD! - Nuxt and Nitro now run the request hooks of evlog plugins. A plugin listed in plugins of evlog.config.ts gets extendLogger and onRequestStart when the request logger is created, and onRequestFinish after the wide event is emitted, with the status, duration and error. Before, those apps ran only a plugin's enrich, drain and keep, so a plugin such as createOtelPlugin() from evlog/otel did nothing there.

    // evlog.config.ts
    import { defineEvlog } from "evlog";
    import { createOtelPlugin } from "evlog/otel";
    
    export default defineEvlog({ plugins: [createOtelPlugin()] });
  • #1099 3f0ba6b Thanks @HugoRCD! - The evlog/nuxt module takes client: false for server-only apps: evlog: { client: false } in nuxt.config.ts skips the browser plugin, the client auto-imports (log, setIdentity, clearIdentity, setMinLevel) and runtimeConfig.public.evlog. The Nitro plugin, the error handler and the server auto-imports (useLogger, log, createEvlogError) stay registered, and createEvlogError and parseError stay auto-imported in app code. The default, true, keeps the current behavior.

  • #1066 cc213da Thanks @HugoRCD! - Add evlog/otel. createOtelPlugin() reads the OpenTelemetry span active when a request starts and sets traceId and spanId on that request's wide event, so the OTLP drain links the log record to its trace.

    import { createOtelPlugin } from "evlog/otel";
    
    app.use(evlog({ plugins: [createOtelPlugin()] }));

    @opentelemetry/api is a new optional peer at >=1.9.0 <2, which covers apps on OpenTelemetry JS SDK 2 and SDK 3. evlog adds no runtime dependency, and an app that does not import evlog/otel does not load @opentelemetry/api. The plugin runs on every integration with a plugins option, and on Nuxt and Nitro from the plugins of evlog.config.ts.

  • #1044 e1e6dfe Thanks @Mystery01092000! - createRequestLogger() now accepts operation in its options and puts it on the wide event, so a task, queue worker or cron job can name itself at creation time: createRequestLogger({ operation: 'billing:sync', requestId }). It is the same field log.fork() sets on a child. Before, the option was not part of RequestLoggerOptions and a cast that passed it was silently dropped.

  • #1045 2d1e756 Thanks @Mystery01092000! - Add sampling.minLevel, a severity floor for wide events. initLogger({ sampling: { minLevel: 'warn' } }) drops trace, debug and info wide events before sampling, from request loggers and from standalone createLogger() loggers such as jobs and workflows, and a sampling.keep rule or keep callback cannot bring them back. error and fatal events, and audit events, always pass this floor (error events still follow sampling.rates; fatal events are always kept). It works everywhere sampling is accepted: initLogger(), evlog.config.ts, the Nuxt and Nitro evlog options, the Vite plugin and Next.js. The top-level minLevel is unchanged and still applies to the log API only.

  • #1112 cc25f54 Thanks @HugoRCD! - evlog/workers: the logger withEvlog passes to the handler now has log.fork(label, fn). Workers has no useLogger(), so fn receives the forked logger as its argument. The child emits its own wide event with operation and _parentRequestId, and the fork is registered with ctx.waitUntil:

    export default withEvlog(async (request, env, ctx, log) => {
      log.fork("send_receipt", async (child) => {
        child.set({ receipt: { orderId } });
        await sendReceipt(orderId);
      });
      return Response.json({ ok: true });
    });

    The toolkit's forkBackgroundLogger now runs without storage, for an integration without AsyncLocalStorage, and passes the forked logger to fn in every case.

Patch Changes

  • #1005 b1a2570 Thanks @HugoRCD! - Split adapter batches at each provider's documented per-request limits. A batch above a limit was sent as one request and rejected (usually 413), and the pipeline dropped it. It is now sent as several requests that each fit:

    • Datadog: 1000 entries and 5 MB uncompressed per request (the byte cap is new).
    • Sentry: 100 logs and 1 MiB of log payload per envelope (the byte cap is new).
    • Axiom: 10,000 events per request.
    • Better Stack: 10 MiB per request.
    • PostHog mode: 'events': request bodies under 20 MB on /batch/.

    Batches within the limits send the same single request as before. Retry behavior is unchanged: the drain path keeps its default retries and the pipeline path sends one attempt unless retries is set. A single event larger than the limit is still sent on its own, so the provider decides what to do with it.

  • #1060 fb3df82 Thanks @orshih6! - auditDiff now reports changes to Date, Set, Map and RegExp fields as a replace (a Date is compared by timestamp, so an equal one emits nothing). Patch paths are now valid RFC 6901 JSON Pointers: ~ and / in keys are escaped as ~0 and ~1, and operations on the root value use '' instead of '/'. redactPaths still matches keys that contain / or ~.

  • #1042 79882a2 Thanks @onmax! - The creditCard built-in no longer masks the digit groups of a UUID. A path such as /api/orders/00000000-0000-4000-8000-000000000000 used to come out as /api/orders/****4000-****0000; it is now left as is. Card numbers written with spaces, hyphens or no separator are still masked to ****1111.

  • #1103 b5ed6cd Thanks @HugoRCD! - A custom error's own toJSON now decides how its data and internal fields are logged. A value it rewrites, such as data: '[REDACTED]', is logged as rewritten, and a value it sets to undefined is not logged. Before, only a key that toJSON left out was hidden, and a rewritten value was logged raw. EvlogError is unchanged: logs keep its internal.

  • #1019 2159dcf Thanks @HugoRCD! - defineEvlogHook() from evlog/eve now reads the v27 session events of eve 0.84 and later, eve 1.0 included. Before, it listened for the v26 names, which those versions refuse, so eve build failed with Hook "hooks/evlog.ts" handles "message.received" was removed in session stream version 27. The hook now registers the v26 names only on eve 0.83 and earlier. A turn now ends on turn.settled and a session on session.ended. Token usage comes from usage.recorded, tool calls from call.requested and call.settled, approvals and sign-ins from interaction.opened and interaction.settled, and the user message from delivery.consumed. The v26 events of eve 0.72 to 0.83 keep working.

    eve 0.84 moved AI SDK runtime context from the step.started instrumentation event to OTel integrations, so defineEvlogInstrumentation() only applies before 0.84. On 0.84 and later, add agent/instrumentation/evlog.ts exporting otelIntegration({ runtimeContext: evlogRuntimeContext }), with otelIntegration from eve/instrumentation/otel. evlogRuntimeContext accepts that input, and returns nothing for a model attempt outside a turn instead of throwing.

  • #1101 84873d5 Thanks @HugoRCD! - The evlog executable runs from a directory whose package.json does not parse. Before, it stopped with ERR_INVALID_PACKAGE_CONFIG before reaching the CLI.

  • #1011 6ab90fd Thanks @adelrodriguez! - Type log.fork() as always present on the request logger that an integration attaches: c.get('log') on Hono, req.log on Express and NestJS, log in Elysia handlers, context.get(loggerContext) on React Router and context.log on oRPC. Before, fork was optional there, so c.get('log').fork(...) failed a strict type check and needed fork! or fork?.. The new ForkableLogger type from evlog describes this logger, for example in SvelteKit's App.Locals. useLogger() still returns AuditableLogger with an optional fork, because inside log.fork() work it returns the forked logger, which has no fork.

  • #1108 af7f8b4 Thanks @HugoRCD! - Fix the fs drain losing events when maxFiles and maxSizePerFile are both set. Once cleanup deleted the day's unsuffixed file (2026-03-14.jsonl), the next write recreated it, and because that file sorts oldest, the same cleanup deleted it straight away, so every later event that day was lost. The drain now appends to the day's newest file and rotates to the next index from there.

  • #1117 66b1ed7 Thanks @HugoRCD! - The AI section of pretty output (ai.tokens, ai.tools, ai.steps, ...) now loads on demand with the pretty-error formatter, so production bundles no longer contain it: about 3.6 kB minified and 1 kB gzipped less for an app that calls initLogger() and log. Pretty output is unchanged once it has loaded. An event with ai logged in the same tick as initLogger() prints as soon as it has, and if the process exits before that, ai prints as a plain field.

  • #1116 99c6610 Thanks @HugoRCD! - The pretty-error formatter (the Why: / Fix: / code-frame block in pretty output) now loads on demand with a dynamic import(), so production bundles no longer contain it: about 3.7 kB minified and 600 B gzipped less for an app that calls initLogger() and log. Pretty output is unchanged once the formatter has loaded. An error logged in the same tick as initLogger() prints as soon as it has, and if the process exits before that, it prints with the error message only.

  • #1115 bfec97f Thanks @HugoRCD! - Framework entries (evlog/hono, evlog/express, evlog/next, evlog/nitro, evlog/nuxt, evlog/workers, ...) and evlog/toolkit no longer load the audit helpers (withAudit, signed, auditDiff, auditOnly, auditEnricher, mockAudit). Without a bundler, each of those entries now loads about 15 kB less from dist/ (about 5 kB gzipped). log.audit() works as before, and the root evlog entry still exports every audit helper.

  • #1072 9aba917 Thanks @HugoRCD! - createDrainPipeline from evlog/pipeline, and the AuditActor and AuditFields types from evlog, show their documentation again in the published types and in editor hovers. Each doc block sat directly on top of another one, so TypeScript attached it to nothing. createDrainPipeline gets its description and usage example back, AuditActor the meaning of each actor field, and AuditFields the description of each reserved audit field.

  • #1035 6a5f8ab Thanks @HugoRCD! - The Nitro plugin (evlog/nitro and evlog/nitro/v2, so the Nuxt module too) no longer clears a drain that another Nitro plugin set with initLogger({ drain }). Nitro does not await plugins, so evlog's plugin finished after a synchronous plugin and reset the global drain. Request events still reached evlog:drain hooks, but standalone loggers (createLogger, or createRequestLogger in a scheduled task or queue worker) drained nowhere. The drain registered before evlog's plugin settles is now kept.

  • #1080 beb0f47 Thanks @HugoRCD! - Widen the optional nitro peer range so that nitro@3.0.261103-rc installs without a peer conflict. Before, npm failed with ERESOLVE and pnpm printed a peer warning when a project used that release candidate with the evlog 3 beta.

  • #1109 44ed576 Thanks @HugoRCD! - The Nitro v2, Nitro v3 and Nuxt integrations now use the incoming x-request-id header as the event's requestId, as every other integration does. Before, they ignored it and generated a UUID (or used cf-ray on Cloudflare, which is now the fallback when no x-request-id is sent).

  • #1098 7f6748a Thanks @HugoRCD! - evlog/nitro (Nitro 3) now emits a streamed response's wide event (text/event-stream, NDJSON, chunked, AI SDK UI streams) once its body ends or the client leaves, instead of when the headers are sent. Fields set with log.set() and errors recorded while the body streams now land on the event instead of being dropped with a log.set() called after the wide event was emitted warning, and the event keeps the status the response was sent with. This uses h3's onDispose, so it needs h3 2.0.1 or later; older Nitro v3 prereleases keep the header-time emit.

  • #1011 6ab90fd Thanks @adelrodriguez! - log.fork(label, fn) now runs fn on a request that evlog does not log. Before, on an Elysia route that exclude or include filters out, or with enabled: false, log had no fork, so log.fork!(...) threw log.fork is not a function and the route answered 500. The work runs and emits no event. With waitUntil, evlog registers the work there, as it does on a logged request. If the work throws, evlog reports the error with console.error and log.fork() still resolves.

  • #1043 c87e32f Thanks @onmax! - parseError() now keeps the why, fix and link an Error carries as its own properties. An error class that sets fix on itself and also has a data property, even undefined, used to come back with fix: undefined. Guidance found in the payload still wins.

  • #1061 8c313b5 Thanks @evlogai! - Sentry logs now carry the reserved user.id attribute when the wide event has a user, the same attribute the official Sentry SDK sets from its user scope. The log item container also carries version: 2, as the Sentry logs protocol recommends. Before, the id only appeared under the userId key.

  • #1015 64bac43 Thanks @HugoRCD! - On Hono, React Router and SvelteKit, a request that evlog does not log now gets a logger that accepts every call and emits nothing, on c.get('log'), context.get(loggerContext) and event.locals.log. This covers a route that include or exclude filters out, and every route after initLogger({ enabled: false }). Before, nothing was attached: c.get('log').set(...) threw a TypeError, the documented Hono onError handler threw as well so the request got no response, and context.get(loggerContext) threw in React Router loaders. log.fork() still runs its work on these requests. useLogger() still throws on them.

@evlog/cli@1.0.0-beta.3

Minor Changes

  • #1102 9ce28df Thanks @HugoRCD! - evlog map, evlog init and evlog doctor recognise Effect 4 apps. map reads HttpRouter.add and HttpRouter.route registrations, credits yield* EvlogLogger and yield* useLogger() as the request logger, and treats routerLayer() as the per-request event. init writes src/evlog.ts with initLoggerLayer and routerLayer built from evlog.config.ts. An app that also uses Next.js, Hono or another framework keeps that framework, and an Effect 3 app (@effect/platform) is not detected. --framework effect forces it.

    The "Could not detect a supported framework" error now lists every framework map supports, Express and Fastify included.

  • #1130 57e83ed Thanks @HugoRCD! - evlog logs --session <id> shows one AI conversation: the events whose gen_ai.conversation.id matches, such as an eve session's turns. evlog logs also shows an event delivered twice under the same idempotencyKey once, so a retried drain write no longer appears as two events.

  • #1075 8ee1a17 Thanks @HugoRCD! - evlog map now records what each entry point's wide event carries. Every route in evlog.map.json and in --json gains an event key with the dotted paths log.set() writes (user.id, cart.items) and the literal action of each log.audit() call. Comparing it between two scans shows which fields a change removes from your production logs. Routes that set nothing have no event key, and the existing keys are unchanged.

Patch Changes

  • #1083 9621ea4 Thanks @HugoRCD! - evlog agents and evlog init no longer delete or duplicate text in an AGENTS.md whose evlog markers are not one clean pair. A file that lost <!-- evlog:end --> got a second block on one run and lost the section below it on the next. Both commands now stop with cli.AGENTS_MARKERS_UNPAIRED, exit 1 and leave the file as it is. A marker quoted in a sentence or a code fence is read as text.

    evlog init also changes in four places:

    • A Nitro config that imports something else from evlog/nitro or evlog/nitro/v3 gets import evlog along with modules: [evlog()]. It got the module entry with no import.
    • The seeded errors.ts and audit.ts parse when a service or a route starts with a digit (_3dShopErrors, _2FA_PASSWORD_CREATED) or a route holds a wildcard.
    • A failed npx skills add exits 1, and the AGENTS.md block no longer points at the skill that was not installed. evlog agents already did both.
    • A file that cannot be written stops the run with cli.WRITE_FAILED, which names the file and the ones written before it, under --json too. evlog agents does the same. Both printed a stack trace and left stdout empty.
  • #1128 e1e33d0 Thanks @HugoRCD! - Two fixes to evlog init and evlog agents.

    • On Windows, evlog init --apps apps/web sets up the app. Workspace app paths were built with backslashes there, so the forward-slash form was rejected as unknown, and the JSON app values and the per-app headings carried backslashes. They use forward slashes on every platform, and --apps apps\web still selects the app.
    • When npx skills add fails, the reported error is the last line it wrote before npm's update notice. Before, a failed first run could report npm notice as the error.
  • #1105 2f94179 Thanks @HugoRCD! - evlog logs exits 1 on an evlog.config it cannot read, like evlog map and evlog config. It exited 2, the code for a misused command. evlog upgrade and --debug are marked experimental in --help and the docs: the upgrade output and the debug case file may change in a 1.x release.

  • #1101 84873d5 Thanks @HugoRCD! - The CLI says what is wrong with the environment it is started in, and its output behaves in a pipe.

    • On Node older than 22, every command stops with @evlog/cli needs Node.js 22.0.0 or later, and this is Node.js 20.20.2 and exit code 1. Before, evlog map failed there with a syntax error about globSync, while doctor, logs and --version still ran on a Node the package does not support. evlog doctor no longer has a failing node check or a cli.NODE_TOO_OLD finding, since it never starts there.
    • Next to an evlog older than the peer range, or with none installed, every command stops with the version it needs, the version and path it found, and exit code 1. Before, Node reported a missing export as a syntax error.
    • A package.json that does not parse in the working directory no longer stops the launcher with ERR_INVALID_PACKAGE_CONFIG.
    • --help prints to stdout, so evlog --help | grep map works. Usage shown with a usage error stays on stderr. Flags are listed the way they are typed (--min-score, -f, --follow), and positional arguments have their own section.
    • Colour follows the stream the report is written to: evlog doctor 2> report.txt writes plain text from a terminal. FORCE_COLOR=0 and TERM=dumb turn colour off, and the header rule uses 24-bit colour only where the terminal supports it.
    • A reader that closes the pipe early, as in evlog logs --json | head, ends the run quietly instead of with an unhandled EPIPE error.
  • #1059 4466f57 Thanks @HugoRCD! - Every command now stops on a flag it does not declare, and on a value flag passed with no value, with exit code 2. evlog map --min-scor 90 used to run with no gate and exit 0, and evlog init --dryrun wrote files. A flag written before the command name is checked too: evlog --dry-run init used to run init for real. A script that carries a mistyped flag today starts failing: fix the spelling.

    The other usage errors exit 2 as well, as the README documents: an unknown command, an invalid --min-score, --framework, --format or --limit, and evlog map <entry> for an entry point that does not exist. An empty --min-score, which is what an unset shell variable expands to, is one of them. A bare --baseline no longer takes the flag after it as its value: evlog map --baseline --min-score 80 and evlog map --baseline --cwd apps/web read both flags. A map error is printed as JSON under --format json, as it already was under --json.

  • #1085 03b63a8 Thanks @HugoRCD! - evlog logs reads a log of any size without its memory growing with it. Every view held each matching event until the end of the run: an 800 MB log took 1.2 GB with --limit 1. A view now keeps only the --limit events it shows, and stats keeps one duration per event for the percentiles. The output of every view is unchanged.

  • #1083 9621ea4 Thanks @HugoRCD! - evlog logs applies every --where when the flag is repeated. Only the last clause applied, so --where path=/a --where n=5 returned events on any path. An empty --where, which is what an unset shell variable expands to, is now a usage error with exit 2.

    An event whose duration, level or service is not text no longer crashes every view, --json included.

    A --where value is a number only when it spells one exactly. zip=123 no longer matches "00123", a 19-digit id no longer matches its neighbours, and amount>5000 no longer matches a field that holds text such as "n/a". Write the value as the field holds it: zip=00123.

  • #1113 6011cd1 Thanks @HugoRCD! - evlog map scores each entry point on its own handler instead of on the whole file.

    • In a file that declares several entry points (Express, Hono and Fastify routes, the methods of a Next.js route.ts, server actions, TanStack Start methods), one instrumented handler used to pass every other one in the file, and each route listed the same event.fields. Each entry point is now read from its handler and the functions, classes and constants of the same file it names. An import such as stripe marks only the handlers that use it as sensitive.
    • request.log.info() in a Fastify app that never registers evlog's plugin no longer passes wide-event: that is Fastify's own logger. req.log and request.log count when the app registers evlog's middleware, or when the call is set, audit or fork. request.evlog is recognised.
    • c.get('log').set() called without binding the logger first is credited in Hono.
    • evlog map --all names each row of a shared file by its route (index.ts · GET /health). evlog map <file> lists the other entry points of the file, and evlog map "POST /orders" inspects one by method and path.

    Scores drop in files where an instrumented route was covering for the others, so the rule-set version moves to 3. --baseline against a map written by an earlier CLI exits 2 with the version mismatch: run evlog map once and commit the new evlog.map.json.

  • #1059 4466f57 Thanks @HugoRCD! - evlog map no longer scores a tree it only partly read.

    • A file with a syntax error, or one the scan cannot read, used to lose its routes, which raised the score. The report now warns <file> failed to parse and counts the file as one failing entry point. With --min-score, map.minScore or --baseline, the run stops with cli.MAP_PARSE_FAILED, exits 1, and leaves evlog.map.json as it was. List the file under map.ignore when it is not app code.
    • Directories named build, dist and coverage are skipped at the project root only, so server/api/build/index.get.ts is scanned.
    • Two entry points with the same method and path, such as get('/') in two routers, get distinct ids. --baseline reported regressions on an unchanged tree when they shared one. The id of every other entry point is unchanged, so a committed map stays comparable, and an entry point is still compared with itself when a second one with its method and path appears or goes. If it also moved to another file in that change, it is compared with the entry point that has its method and path and lost the most checks, then the lowest-scoring one.
    • .js, .mjs and .cjs files are parsed with JSX, so a Next.js page.js is checked instead of reported as n/a.

    The first and the last of these can change a verdict on code you did not touch, so the rule-set version moves to 2. --baseline against a map written by an earlier CLI exits 2 with the version mismatch: run evlog map once and commit the new evlog.map.json.

  • #1125 4b2f220 Thanks @HugoRCD! - evlog telemetry status --help (and the other telemetry subcommands) now shows the CLI version in the header and the full evlog telemetry status command in USAGE, like every other command.

  • #1059 4466f57 Thanks @HugoRCD! - New telemetry.fail(code?) for a command that handles its own failure: it prints the error, sets an exit code and returns, so the run used to be recorded as outcome: "success". Call it where the error is handled and the event carries outcome: "error" with code as its errorCode.

    The evlog CLI calls it for every error it reports, so evlog logs with no log directory, evlog config on an invalid file and a mistyped flag are sent as errors with their cli.* code. A failed --min-score or --baseline gate stays a successful run with mapGateFailed: true.

  • #1059 4466f57 Thanks @HugoRCD! - Telemetry now does what its docs say.

    • The run that opts out is no longer queued: telemetry disable, or disableTelemetry() called inside a run, used to purge the outbox and then write its own event to it. disableTelemetry() turns off every handle created for the tool, and a flush() already sending stops at the batch in flight.
    • EVLOG_TELEMETRY_DEBUG=1 prints each run's event as the run ends and makes no request. The event is not queued either. It used to print at the next flush, and that flush sent the events.
    • An endpoint that never answers no longer keeps the process alive for ten seconds. The drain closes its connection at the 400 ms timeout. It no longer follows redirects, so endpoint has to be the final URL.
    • DO_NOT_TRACK=true, yes and on opt out like 1, and EVLOG_TELEMETRY=false, no and off like 0.
    • withTelemetry() records only the flags a command declares. An undeclared flag used to be sent by name, and a positional as <set>.
    • New bin option: the command people type, when it is not name. The first-run notice prints <bin> telemetry disable, so the evlog CLI's notice now says evlog telemetry disable, not evlog-cli telemetry disable.
    • evlog telemetry status lists the allowlisted init*, map* and logs* values. The fields evlog logs records are documented.
  • Updated dependencies [777e3bb, 4466f57, 4466f57]:

    • @evlog/telemetry@1.0.0-beta.1

@evlog/telemetry@1.0.0-beta.1

Minor Changes

  • #1052 777e3bb Thanks @HugoRCD! - run() writes every event to the on-disk outbox and POSTs only when the last successful flush is older than 15 minutes (60 minutes in CI) or 20 events have queued, capped at 50 per POST. Failed uploads stay on disk for the next command. flush() still sends immediately. Override the window with flushIntervalMs or EVLOG_TELEMETRY_FLUSH_INTERVAL_MS. createGitHubActionsTelemetry() keeps flushing after every run(), since runners discard the outbox when the job ends.

Patch Changes

  • #1059 4466f57 Thanks @HugoRCD! - New telemetry.fail(code?) for a command that handles its own failure: it prints the error, sets an exit code and returns, so the run used to be recorded as outcome: "success". Call it where the error is handled and the event carries outcome: "error" with code as its errorCode.

    The evlog CLI calls it for every error it reports, so evlog logs with no log directory, evlog config on an invalid file and a mistyped flag are sent as errors with their cli.* code. A failed --min-score or --baseline gate stays a successful run with mapGateFailed: true.

  • #1059 4466f57 Thanks @HugoRCD! - Telemetry now does what its docs say.

    • The run that opts out is no longer queued: telemetry disable, or disableTelemetry() called inside a run, used to purge the outbox and then write its own event to it. disableTelemetry() turns off every handle created for the tool, and a flush() already sending stops at the batch in flight.
    • EVLOG_TELEMETRY_DEBUG=1 prints each run's event as the run ends and makes no request. The event is not queued either. It used to print at the next flush, and that flush sent the events.
    • An endpoint that never answers no longer keeps the process alive for ten seconds. The drain closes its connection at the 400 ms timeout. It no longer follows redirects, so endpoint has to be the final URL.
    • DO_NOT_TRACK=true, yes and on opt out like 1, and EVLOG_TELEMETRY=false, no and off like 0.
    • withTelemetry() records only the flags a command declares. An undeclared flag used to be sent by name, and a positional as <set>.
    • New bin option: the command people type, when it is not name. The first-run notice prints <bin> telemetry disable, so the evlog CLI's notice now says evlog telemetry disable, not evlog-cli telemetry disable.
    • evlog telemetry status lists the allowlisted init*, map* and logs* values. The fields evlog logs records are documented.

@github-actions
github-actions Bot requested a review from HugoRCD as a code owner November 7, 2026 15:10
@vercel

vercel Bot commented Nov 7, 2026 •

Copy link
Copy Markdown
Contributor

The latest updates on your projects. Learn more about Vercel for GitHub.

5 Skipped Deployments
Project Deployment Actions Updated
evi Skipped Skipped Nov 16, 2026 9:29pm UTC
evlog-docs Skipped Skipped Nov 16, 2026 9:29pm UTC
evlog-render-lab Skipped Skipped Nov 16, 2026 9:29pm UTC
evlog-telemetry Skipped Skipped Nov 16, 2026 9:29pm UTC
just-use-evlog Skipped Skipped Nov 16, 2026 9:29pm UTC

Request Review

@coderabbitai

coderabbitai Bot commented Nov 7, 2026 •

Copy link
Copy Markdown
Contributor

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

Run configuration

  • Source mode: Auto
  • Repository: Unavailable
  • Branch: Unavailable
  • File: No configuration file from this repository applied
  • Commit: Unavailable
  • Layers: defaults
  • Review profile: CHILL
  • Plan: Advanced
  • Run ID: 4b9a6fa9-4802-4465-b158-92ba0e57d9a8

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • Trigger review
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

@github-actions
github-actions Bot force-pushed the changeset-release/v3 branch 3 times, most recently from 7ddc96a to cc7032d Compare November 14, 2026 10:10
@HugoRCD
HugoRCD force-pushed the changeset-release/v3 branch from cc7032d to 9586063 Compare November 14, 2026 12:45
@github-actions

Copy link
Copy Markdown
Contributor Author

Thank you for following the naming conventions! 🙏

@pkg-pr-new

pkg-pr-new Bot commented Nov 14, 2026 •

Copy link
Copy Markdown
@evlog/cli

npm i https://pkg.pr.new/@evlog/cli@1017

evlog

npm i https://pkg.pr.new/evlog@1017

@evlog/nuxthub

npm i https://pkg.pr.new/@evlog/nuxthub@1017

@evlog/signals

npm i https://pkg.pr.new/@evlog/signals@1017

@evlog/telemetry

npm i https://pkg.pr.new/@evlog/telemetry@1017

commit: 0a26547

@vercel
vercel Bot temporarily deployed to Preview – evlog-docs November 15, 2026 19:07 Inactive
@vercel
vercel Bot temporarily deployed to Preview – evlog-render-lab November 15, 2026 19:07 Inactive
@vercel
vercel Bot temporarily deployed to Preview – just-use-evlog November 15, 2026 19:07 Inactive
@vercel
vercel Bot temporarily deployed to Preview – evlog-telemetry November 15, 2026 19:07 Inactive
@vercel
vercel Bot temporarily deployed to Preview – evi November 15, 2026 19:07 Inactive
@github-actions

github-actions Bot commented Nov 15, 2026 •

Copy link
Copy Markdown
Contributor Author

Risk score: Typical (2/4)

Assessed at 0a26547 from the changed files and the title, by scripts/risk-score. A new push replaces this score.

  • On merge: publishes evlog 3.0.0-beta.3 (prerelease)
  • evlog core: Release: publishes 3.0.0-beta.3 (prerelease): evlog/package.json
  • CI and release: Release: records changesets for the prerelease: .changeset/pre.json
  • 3 files changed (49 lines), 0 of them tests
Area Level Why
evlog core Typical Release: publishes 3.0.0-beta.3 (prerelease): evlog/package.json; Docs or changelog: evlog/CHANGELOG.md
CI and release Typical Release: records changesets for the prerelease: .changeset/pre.json

@HugoRCD
HugoRCD force-pushed the changeset-release/v3 branch from 0789787 to 9331f21 Compare November 16, 2026 17:53
@vercel
vercel Bot temporarily deployed to Preview – evlog-docs November 16, 2026 17:53 Inactive
@vercel
vercel Bot temporarily deployed to Preview – evlog-telemetry November 16, 2026 17:53 Inactive
@vercel
vercel Bot temporarily deployed to Preview – just-use-evlog November 16, 2026 17:53 Inactive
@vercel
vercel Bot temporarily deployed to Preview – evlog-render-lab November 16, 2026 17:53 Inactive
@vercel
vercel Bot temporarily deployed to Preview – evi November 16, 2026 17:53 Inactive
@github-actions
github-actions Bot force-pushed the changeset-release/v3 branch 4 times, most recently from 75b2aea to d3be3c0 Compare November 16, 2026 21:25
@vercel
vercel Bot temporarily deployed to Preview – evlog-render-lab November 16, 2026 21:29 Inactive
@vercel
vercel Bot temporarily deployed to Preview – evlog-docs November 16, 2026 21:29 Inactive
@vercel
vercel Bot temporarily deployed to Preview – just-use-evlog November 16, 2026 21:29 Inactive
@github-actions
github-actions Bot force-pushed the changeset-release/v3 branch from e59cc82 to 4ecd873 Compare November 17, 2026 18:41
@github-actions
github-actions Bot force-pushed the changeset-release/v3 branch from 4ecd873 to 1daf336 Compare November 17, 2026 18:46
@github-actions
github-actions Bot force-pushed the changeset-release/v3 branch from 1daf336 to 68820ed Compare November 17, 2026 18:50
@github-actions
github-actions Bot force-pushed the changeset-release/v3 branch from 68820ed to 7c632c6 Compare November 17, 2026 18:52

This branch was successfully deployed

5 active deployments
Preview – evlog-telemetry — 01190900 Deployed Nov 18, 2026 by vercel[bot]
Preview – evi — 01190900 Deployed Nov 18, 2026 by vercel[bot]
Preview – evlog-docs — 01190900 Deployed Nov 18, 2026 by vercel[bot]
Preview – evlog-render-lab — 01190900 Deployed Nov 18, 2026 by vercel[bot]
Preview – just-use-evlog — 01190900 Deployed Nov 18, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

0 participants