[GrowthX-Team]

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

Releases

@outputai/core@0.16.0

Minor Changes

• 2156c4b: Folded @outputai/llm as a feature of @outputai/core, under the @outputai/core/llm entry point. Usage is the same: replace @outputai/llm imports with @outputai/core/llm. The @outputai/llm package is no longer published. Other changes are:

  • Added ai as a peer dependency of @outputai/core, and the built-in AI SDK providers (@ai-sdk/amazon-bedrock, @ai-sdk/anthropic, @ai-sdk/azure, @ai-sdk/google-vertex, @ai-sdk/openai and @ai-sdk/perplexity) as optional peer dependencies. Install the providers your prompts use; using a provider whose package is not installed throws a FatalError naming the package to install.
  • Changed getProviderNames() to list only the built-in providers whose package is installed, plus registered ones.
  • Updated the CLI project templates and agent instructions to use @outputai/core/llm, and added @ai-sdk/anthropic and @ai-sdk/openai to the project template.

  The llm:generation:metering and cost:llm:request events and the LLM trace attributes are unchanged.

• 91027b1: Added public type exports for HTTP fetch, workflow definitions, credential provider contexts, LLM calls and eval verification.

• c2756ea: Added per-request context-tier pricing for LLM models, including tier-specific input, output, cache-read and cache-write rates. Applied tiers are exposed on normalized cost items, and multi-step calls are priced separately per provider request.

• 755b932: Added usage and cost tracking for Anthropic and OpenAI provider-native web search, plus Google Search grounding pricing for Gemini 4. Each billed search is recorded as a tools item, priced separately from tokens, and included in the normalized generation cost total.

• 69acd83: Stopped publishing the @outputai/output umbrella package. Projects depend on @outputai/core and, as a dev dependency, @outputai/cli instead. Other changes are:

  • Changed the project template to depend on @outputai/core and @outputai/cli instead of @outputai/output.
  • Changed output fix to replace an @outputai/output dependency with @outputai/core in dependencies and @outputai/cli in devDependencies, at the same version.
  • Changed the output update warning about a version range that holds back the CLI so it no longer refers to @outputai/output.
  • Removed the internal @outputai/core/sdk/helpers and @outputai/core/sdk/runtime entry points.
  • Removed support for the output-workflow-bundle package export condition when bundling workflows.

• e254388: Folded @outputai/http as a feature of @outputai/core, under the @outputai/core/http entry point. Usage is the same: replace @outputai/http imports with @outputai/core/http. The @outputai/http package is no longer published. Other changes are:

  • Added ky and undici as peer dependencies of @outputai/core, and removed undici from its regular dependencies.
  • Added HTTPRequestCount and HTTPRequestCost to @outputai/core/http.
  • Updated the CLI project templates and agent instructions to use @outputai/core/http.
  • Changed verbose HTTP tracing (OUTPUT_TRACE_HTTP_VERBOSE) to record request and response bodies with the same parsing as sendHttpRequest(), so multipart bodies are now stored as base64.

  Changed sendHttpRequest() response body parsing when responseOptions.includeBody is true:

  • Changed XML, form-urlencoded and JavaScript content types to be returned as text instead of base64.
  • Changed empty or invalid JSON bodies to be returned as raw text instead of throwing.

• 5db73d3: Folded @outputai/evals as a feature of @outputai/core, under the @outputai/core/evals entry point. Replace @outputai/evals imports with @outputai/core/evals, except for the LLM judges, which moved to @outputai/core/llm. The @outputai/evals package is no longer published. Other changes are:

  • Moved judgeVerdict, judgeScore, judgeBoolean, judgeLabel and the JudgeArgs type to @outputai/core/llm. Their arguments and results are unchanged.
  • Removed interpretResult, aggregateCaseVerdict, renderEvalOutput, computeExitCode, getEvalWorkflowName, isEvalWorkflow and getParentWorkflowName from the public API. They were used internally by the eval workflow and the CLI.
  • Removed the internal Zod schemas from the public API. DatasetSchema and EvalOutputSchema are still exported, along with the VERDICT and CRITICALITY values and the eval types.
  • Changed the workflow bundler to replace steps and evaluators that are referenced as values in workflow files, rather than called, with their activity names, and to strip their imports. Previously their modules were bundled into the workflow sandbox when the file had no activity calls.
  • Removed @outputai/evals from @outputai/cli.

• fd6bb62: Added worker memory sizing diagnostics and deferred the workflow concurrency defaults to @temporalio/worker.

  Worker sizing startup log

  Added a Worker sizing record, logged once at startup with the thread pool size, reuseV8Context, the isolate count, the per-isolate heap limit, the heap all isolates could commit together, the host's memory limit, the resolved cache size, and the workflow task slot count.

  Warns when the projected heap exceeds the host's memory limit, and when there are more workflow task slots than cache entries, which causes tasks to evict each other's workflows and forces history replays.

  Accounts for NODE_OPTIONS heap caps applying per isolate rather than per process: a worker runs one V8 isolate on its main thread plus one per workflow thread, so the heap it can commit is a multiple of the cap.

  Periodic telemetry

  • Added memory.nonMainHeap (rss - heapUsed): resident memory outside the main isolate's live objects, including workflow thread heaps, Rust core allocations, and heap pages V8 has committed but not filled. Cached workflow memory is counted here.
  • Added memory.heapSizeLimit, the per-isolate heap ceiling, to each sample.
  • Renamed memory.constrainedMemory to memory.memoryLimit and changed what it reports. process.constrainedMemory() returns a sentinel near 2^64 rather than 0 when the process is unconstrained, so the old field was not usable outside a memory-capped container. memoryLimit falls back to the machine's total memory. Anything parsing the old field needs updating.

  Environment variables

  • Added TEMPORAL_WORKFLOW_THREAD_POOL_SIZE, the number of threads running workflow sandboxes, which sets the number of isolates. Unset by default.
  • Removed the TEMPORAL_MAX_CACHED_WORKFLOWS default of 1000. When unset, the cache is sized from the isolate's heap limit instead of a fixed value. That formula assumes roughly 600 cached workflows per GB, so deployments with heavier payloads should set it explicitly.
  • Removed the TEMPORAL_MAX_CONCURRENT_WORKFLOW_TASK_EXECUTIONS default of 200. When unset, it falls back to the upstream default of 40.

Patch Changes

• 17f6c8d: Removed the duplicated cause from step and workflow failures. The ApplicationFailure cause is now the error's own .cause instead of the error itself, so its message and stack are no longer encoded twice. As a result, wherever a failure's cause chain is exposed, such as API error responses, traces and logs, it is now one redundant level shorter.

• 606278c: Reduce worker memory usage by loading only the invoked workflow module into each cached Temporal workflow execution.

• eaacecf: Dependency updates:

  • @hono/node-server: 2.0.10 -> 2.1.3, reached through @modelcontextprotocol/sdk
  • @modelcontextprotocol/sdk: 1.31.0 -> 1.32.0, reached through @anthropic-ai/claude-agent-sdk
  • body-parser: 2.3.0 -> 2.4.0, reached through express
  • fast-uri: 3.1.8 -> 3.1.9, reached through ajv
  • hono: 4.13.10 -> 4.13.11, reached through @modelcontextprotocol/sdk
  • ip-address: 10.7.2 -> 10.7.4, reached through @modelcontextprotocol/sdk
  • qs: 6.16.0 -> 6.16.1, reached through express and body-parser
  • undici: 7.29.1 -> 7.30.1, reached through @ai-sdk/provider-utils
  • undici: 8.11.0 -> 8.12.3, fixing Node's built-in fetch returning gzip response bodies undecoded after Undici is imported
  • zod: 4.6.5 -> 4.6.6

  Peer dependency ranges:

  • undici min supported peer dependency version for @outputai/core is now 8.12.2.
  • zod min supported peer dependency version for @outputai/core is now 4.6.6.

@outputai/cli@0.16.0

Patch Changes

• 2156c4b: Folded @outputai/llm as a feature of @outputai/core, under the @outputai/core/llm entry point. Usage is the same: replace @outputai/llm imports with @outputai/core/llm. The @outputai/llm package is no longer published. Other changes are:

  • Added ai as a peer dependency of @outputai/core, and the built-in AI SDK providers (@ai-sdk/amazon-bedrock, @ai-sdk/anthropic, @ai-sdk/azure, @ai-sdk/google-vertex, @ai-sdk/openai and @ai-sdk/perplexity) as optional peer dependencies. Install the providers your prompts use; using a provider whose package is not installed throws a FatalError naming the package to install.
  • Changed getProviderNames() to list only the built-in providers whose package is installed, plus registered ones.
  • Updated the CLI project templates and agent instructions to use @outputai/core/llm, and added @ai-sdk/anthropic and @ai-sdk/openai to the project template.

  The llm:generation:metering and cost:llm:request events and the LLM trace attributes are unchanged.

• 69acd83: Stopped publishing the @outputai/output umbrella package. Projects depend on @outputai/core and, as a dev dependency, @outputai/cli instead. Other changes are:

  • Changed the project template to depend on @outputai/core and @outputai/cli instead of @outputai/output.
  • Changed output fix to replace an @outputai/output dependency with @outputai/core in dependencies and @outputai/cli in devDependencies, at the same version.
  • Changed the output update warning about a version range that holds back the CLI so it no longer refers to @outputai/output.
  • Removed the internal @outputai/core/sdk/helpers and @outputai/core/sdk/runtime entry points.
  • Removed support for the output-workflow-bundle package export condition when bundling workflows.

• e254388: Folded @outputai/http as a feature of @outputai/core, under the @outputai/core/http entry point. Usage is the same: replace @outputai/http imports with @outputai/core/http. The @outputai/http package is no longer published. Other changes are:

  • Added ky and undici as peer dependencies of @outputai/core, and removed undici from its regular dependencies.
  • Added HTTPRequestCount and HTTPRequestCost to @outputai/core/http.
  • Updated the CLI project templates and agent instructions to use @outputai/core/http.
  • Changed verbose HTTP tracing (OUTPUT_TRACE_HTTP_VERBOSE) to record request and response bodies with the same parsing as sendHttpRequest(), so multipart bodies are now stored as base64.

  Changed sendHttpRequest() response body parsing when responseOptions.includeBody is true:

  • Changed XML, form-urlencoded and JavaScript content types to be returned as text instead of base64.
  • Changed empty or invalid JSON bodies to be returned as raw text instead of throwing.

• eaacecf: Dependency updates:

  • @hono/node-server: 2.0.10 -> 2.1.3, reached through @modelcontextprotocol/sdk
  • @modelcontextprotocol/sdk: 1.31.0 -> 1.32.0, reached through @anthropic-ai/claude-agent-sdk
  • body-parser: 2.3.0 -> 2.4.0, reached through express
  • fast-uri: 3.1.8 -> 3.1.9, reached through ajv
  • hono: 4.13.10 -> 4.13.11, reached through @modelcontextprotocol/sdk
  • ip-address: 10.7.2 -> 10.7.4, reached through @modelcontextprotocol/sdk
  • qs: 6.16.0 -> 6.16.1, reached through express and body-parser
  • undici: 7.29.1 -> 7.30.1, reached through @ai-sdk/provider-utils
  • undici: 8.11.0 -> 8.12.3, fixing Node's built-in fetch returning gzip response bodies undecoded after Undici is imported
  • zod: 4.6.5 -> 4.6.6

  Peer dependency ranges:

  • undici min supported peer dependency version for @outputai/core is now 8.12.2.
  • zod min supported peer dependency version for @outputai/core is now 4.6.6.

• 4a437bf: Patching handlebars version from 4.7.9 to 4.7.10 to fix vulnerabilities.

• 5db73d3: Folded @outputai/evals as a feature of @outputai/core, under the @outputai/core/evals entry point. Replace @outputai/evals imports with @outputai/core/evals, except for the LLM judges, which moved to @outputai/core/llm. The @outputai/evals package is no longer published. Other changes are:

  • Moved judgeVerdict, judgeScore, judgeBoolean, judgeLabel and the JudgeArgs type to @outputai/core/llm. Their arguments and results are unchanged.
  • Removed interpretResult, aggregateCaseVerdict, renderEvalOutput, computeExitCode, getEvalWorkflowName, isEvalWorkflow and getParentWorkflowName from the public API. They were used internally by the eval workflow and the CLI.
  • Removed the internal Zod schemas from the public API. DatasetSchema and EvalOutputSchema are still exported, along with the VERDICT and CRITICALITY values and the eval types.
  • Changed the workflow bundler to replace steps and evaluators that are referenced as values in workflow files, rather than called, with their activity names, and to strip their imports. Previously their modules were bundled into the workflow sandbox when the file had no activity calls.
  • Removed @outputai/evals from @outputai/cli.

• Updated dependencies [2156c4b]

• Updated dependencies [17f6c8d]

• Updated dependencies [91027b1]

• Updated dependencies [c2756ea]

• Updated dependencies [755b932]

• Updated dependencies [606278c]

• Updated dependencies [69acd83]

• Updated dependencies [e254388]

• Updated dependencies [eaacecf]

• Updated dependencies [5db73d3]

• Updated dependencies [fd6bb62]

  • @outputai/core@0.16.0

output-api@0.16.0

Patch Changes

• 17f6c8d: Removed the duplicated cause from step and workflow failures. The ApplicationFailure cause is now the error's own .cause instead of the error itself, so its message and stack are no longer encoded twice. As a result, wherever a failure's cause chain is exposed, such as API error responses, traces and logs, it is now one redundant level shorter.

• eaacecf: Dependency updates:

  • @hono/node-server: 2.0.10 -> 2.1.3, reached through @modelcontextprotocol/sdk
  • @modelcontextprotocol/sdk: 1.31.0 -> 1.32.0, reached through @anthropic-ai/claude-agent-sdk
  • body-parser: 2.3.0 -> 2.4.0, reached through express
  • fast-uri: 3.1.8 -> 3.1.9, reached through ajv
  • hono: 4.13.10 -> 4.13.11, reached through @modelcontextprotocol/sdk
  • ip-address: 10.7.2 -> 10.7.4, reached through @modelcontextprotocol/sdk
  • qs: 6.16.0 -> 6.16.1, reached through express and body-parser
  • undici: 7.29.1 -> 7.30.1, reached through @ai-sdk/provider-utils
  • undici: 8.11.0 -> 8.12.3, fixing Node's built-in fetch returning gzip response bodies undecoded after Undici is imported
  • zod: 4.6.5 -> 4.6.6

  Peer dependency ranges:

  • undici min supported peer dependency version for @outputai/core is now 8.12.2.
  • zod min supported peer dependency version for @outputai/core is now 4.6.6.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
outputai	🟢 Ready	View Preview	Nov 2, 2026, 3:00 PM

💡 Tip: Enable Automations to automatically generate PRs for you.

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): This release bumps the OUTPUT_API_VERSION default in sdk/cli/src/assets/docker/docker-compose-dev.yml to 0.15.1, but the env var table in docs/guides/packages/cli.mdx:826 still documents the default as 0.12.0. The release pipeline (ops/bump_release.sh → set_sdk_version.js) doesn't touch that page, so the gap widens every release — worth either updating it or dropping the hardcoded default from the table.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

None.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): This PR bumps the compose default to outputai/api:0.16.0 (sdk/cli/src/assets/docker/docker-compose-dev.yml:83), but the docs that hardcode that tag still say 0.12.0 — docs/guides/packages/cli.mdx:826 (OUTPUT_API_VERSION default) and the Railway/Render deployment guides. The release pipeline (ops/bump_release.sh → set_sdk_version.js) only rewrites the compose file, so these drift one release further every time. Worth either updating them or teaching the script about them.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): docs/guides/packages/cli.mdx:826 documents OUTPUT_API_VERSION as defaulting to 0.12.0, but this PR moves the real default in sdk/cli/src/assets/docker/docker-compose-dev.yml to 0.16.0. Nothing in ops/bump_release.sh / set_sdk_version.js updates that table, so the gap widens every release. Consider replacing the hardcoded value with "matches the installed CLI version".

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

None.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): The release only rewrites the API image version in sdk/cli/src/assets/docker/docker-compose-dev.yml, so the self-host deployment guides keep drifting: docs/guides/operations/deployment/render.mdx:99 and railway.mdx:47 still tell users to deploy docker.io/outputai/api:0.12.0, now four minors behind. Consider adding those pinned references to sdk/cli/scripts/set_sdk_version.js so each release bumps them too.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Compatibility): The release ships "Reverted undici from 8.11.0 to 8.10.2" as a fix, but undici is only a peer dependency of @outputai/core with range >=8 <9 (sdk/core/package.json), and 8.10.2 is pinned only in the monorepo catalog. A fresh consumer install resolves the latest 8.x — including the 8.11.0 the changeset says breaks gzip for plain fetch/AI SDK providers. The v0.15.0→v0.16.0 guide even instructs npm install undici@8. Consider narrowing the peer range (e.g. >=8 <8.11.0) or calling the constraint out in the guide.

2. Nice-to-have (Documentation): The generated v0.16.0 changelog discloses that the ApplicationFailure cause chain is now one level shorter "wherever a failure's cause chain is exposed, such as API error responses, traces and logs" — a user-observable response-shape change — but the linked docs/guides/migrations/v0.15.0-to-v0.16.0.mdx never mentions it. Worth one bullet there plus a checklist line for anyone parsing API error cause nesting.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): The release bumps the OUTPUT_API_VERSION default in sdk/cli/src/assets/docker/docker-compose-dev.yml to 0.16.0, but docs/guides/packages/cli.mdx:826 still documents that default as 0.12.0 (and operations/deployment/{railway,render}.mdx still pin outputai/api:0.12.0). Pre-existing drift, but it widens every release — worth folding into whatever regenerates the version-bearing files.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

None.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Compatibility): The brave-otters-wander change (ApplicationFailure cause is now the error's own .cause, shortening the cause chain in API error responses) is a user-visible contract change but appears in neither docs/guides/migrations/v0.15.0-to-v0.16.0.mdx nor api/CHANGELOG.md (which gets an empty ## 0.16.0). Callers walking cause.cause in API error payloads get no upgrade note outside the general changelog.

2. Nice-to-have (Documentation): In docs/guides/changelog/index.mdx:23 the sentence "The llm:generation:metering and cost:llm:request events ... are unchanged." has no blank line before it, so Mintlify renders it as a continuation of the preceding bullet instead of its own paragraph. Same text in releases.json summary for amber-hounds-glide; fixing it there and re-running ./ops/regenerate_docs.sh cleans both.

3. Nice-to-have (Quality): changeset version re-expanded the hand-compacted peerDependenciesMeta entries in sdk/core/package.json ({ "optional": true } → multi-line), adding unrelated churn to the release diff. This will recur every release unless the compact form is dropped or ops/bump_release.sh runs the formatter over bumped package.json files.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Compatibility): The 17f6c8d entry ships a user-visible change to serialized failures ("wherever a failure's cause chain is exposed, such as API error responses, traces and logs, it is now one redundant level shorter"), but docs/guides/migrations/v0.15.0-to-v0.16.0.mdx never mentions it. Consumers walking error.cause.cause on /workflow/run responses get a different shape with no upgrade note. A bullet in "What changed" would close it.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Compatibility): The 17f6c8d change ("wherever a failure's cause chain is exposed, such as API error responses, traces and logs, it is now one redundant level shorter") ships in this release but is absent from docs/guides/migrations/v0.15.0-to-v0.16.0.mdx and its checklist. Consumers that walk cause.cause on API errors or trace failures will silently read a different shape. Worth a short section/checklist line there.

2. Nice-to-have (Documentation): output-api@0.16.0 gets an empty ## 0.16.0 section in api/CHANGELOG.md and no mention in docs/guides/changelog/index.mdx, even though the image's error-response shape changes via the bundled core patch. Fixed-group bumps with no own changeset are expected, but an API-user-visible change landing with a blank API changelog is easy to miss.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[claude]

Nice-to-have / Documentation: this sentence has no blank line above it, so CommonMark lazy-continues it into the preceding bullet — it renders as part of "Updated the CLI project templates…" instead of a standalone paragraph. Same artifact lands in sdk/core/CHANGELOG.md and the amber-hounds-glide summary in releases.json.

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): In the generated changelog, docs/guides/changelog/index.mdx:23 ("The llm:generation:metering and cost:llm:request events … are unchanged.") has no blank line above it, so Markdown lazy-continuation folds it into the preceding bullet instead of rendering it as its own paragraph. Same artifact appears in sdk/core/CHANGELOG.md and the amber-hounds-glide summary in releases.json; fixing the summary in releases.json and re-running ./ops/regenerate_docs.sh clears the docs page.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): In docs/guides/changelog/index.mdx:23 the sentence "The llm:generation:metering and cost:llm:request events and the LLM trace attributes are unchanged." sits directly under the preceding bullet with no blank line, so Markdown lazy-continuation folds it into the last list item instead of rendering it as its own paragraph. The generated region can't be hand-edited — fix the amber-hounds-glide summary in docs/guides/data/releases.json (add \n\n before that sentence) and re-run ./ops/regenerate_docs.sh.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[claude]

Nice-to-have (Quality): changeset version reflowed the hand-compacted peerDependenciesMeta one-liners into expanded objects. Harmless, but it will recur on every release PR unless ops/bump_release.sh formats package.json afterwards (or main adopts the expanded form).

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Quality): changeset version reflowed the hand-compacted peerDependenciesMeta entries in sdk/core/package.json:95 into expanded objects — pure formatting churn that will recur on every release PR. Consider running ops/format_files.sh (or a JSON formatter) at the end of ops/bump_release.sh, or just accept the expanded form on main.

2. Nice-to-have (Documentation): The release bumps every in-repo version reference the sync script knows about (framework_version.json, docker-compose-dev.yml), but docs/guides/operations/deployment/railway.mdx:47 and render.mdx:99 still tell users to deploy docker.io/outputai/api:0.12.0 — now four minors behind. Worth folding those into the release bump or switching them to a non-pinned reference.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

1. Nice-to-have (Documentation): The release pipeline bumps the API image default in docker-compose-dev.yml to 0.16.0, but the self-hosting guides still instruct users to deploy docker.io/outputai/api:0.12.0 (docs/guides/operations/deployment/railway.mdx:47, render.mdx:99) — now four minors behind the worker/SDK this release publishes. Consider having set_sdk_version.js rewrite those pinned tags too, or switch the guides to a floating/latest-style reference.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS

[github-actions]

PR review

Verdict

✅ PASS

Findings

None.

Categories

• Design: ✅ PASS
• Quality: ✅ PASS
• Correctness: ✅ PASS
• Documentation: ✅ PASS
• Changeset: ✅ PASS
• Tests: ✅ PASS
• Security: ✅ PASS
• Compatibility: ✅ PASS