feat(api): update API spec from langfuse/langfuse 8ff680b

[langfuse-bot]

Confidence Score: 4/5

The PR should not merge until the still-relevant sdk-log import contract and the explicit import-placement requirement are addressed.

Summary

The PR refreshes generated API models and documentation, adds sync and async unstable skills operations, extends several request parameters, and consolidates cursor metadata.

• The skills service supports listing, version creation, retrieval, updates, file-content reads, labeling, and deletion.
• Legacy ingestion guidance is updated, while its typed sdk-log import surface is removed.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    API[Langfuse API client] --> U[Unstable client]
    U --> S[Skills client]
    S --> R[Shared HTTP transport]
    R --> E[Skills API]
    E --> M[Typed skill models]

Loading

_{Reviews (1) · Last reviewed commit: "feat(api): update API spec from langfuse..."}

[github-actions]

@claude review

[claude]

Claude Code Review

No review was started: this request came from a bot account. Manual reviews can only be requested by someone with write access to this repository. Ask a maintainer to comment @claude review, or have your automation post the comment from a user account with write access.

_{Tip: disable this comment in your organization's Code Review settings.}

[greptile-apps]

Removed supported sdk-log models The batch documentation says non-score events remain accepted on Cloud until November 16, 2026, and on self-hosted deployments until v4-only write mode is enabled. This change removes sdk-log from the typed event union and removes its public models, so existing callers importing SdkLogBody, SdkLogEvent, or IngestionEvent_SdkLog fail before they can submit those events.

Knowledge Base Used: API client and service surface

Prompt To Fix With AI

This is a comment left during a code review.
Path: langfuse/api/ingestion/types/ingestion_event.py
Line: 134

Comment:
**Removed supported sdk-log models** The batch documentation says non-score events remain accepted on Cloud until November 16, 2026, and on self-hosted deployments until v4-only write mode is enabled. This change removes `sdk-log` from the typed event union and removes its public models, so existing callers importing `SdkLogBody`, `SdkLogEvent`, or `IngestionEvent_SdkLog` fail before they can submit those events.

**Knowledge Base Used:** [API client and service surface](https://app.greptile.com/personal-org-4986/-/custom-context/knowledge-base/langfuse/langfuse-python/-/docs/api-client-and-service-surface.md)

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

[claude]

I reviewed this PR and didn't find any bugs. Because it's a large, fully auto-generated change spanning 68 files (a new "unstable/skills" API resource plus a new generic cursor-pagination type, alongside smaller additive changes to existing clients), a human look would still be worthwhile to confirm the regenerated client matches the intended upstream Fern/OpenAPI spec.

What was reviewed:

• New unstable/skills client, raw client, and ~14 type modules for skill versions, files, and list pagination
• New CursorMetaResponse generic pagination type in langfuse/api/utils/pagination
• Type narrowing of Dataset/CreateDatasetRequest inputSchema/expectedOutputSchema from Any to Dict[str, Any] — applied consistently to both request and response models, consistent with an intentional spec tightening rather than an isolated regression
• Additive/renaming changes across comments, experiments, ingestion events, observations, opentelemetry, scores, trace, legacy observations_v1, metrics, and prompts clients

Extended reasoning...

The diff touches only auto-generated code under langfuse/api/ (Fern codegen from an upstream API spec bump), adding a new unstable "skills" resource and a generic cursor-pagination type plus smaller additive/renaming changes to existing typed clients; there is no hand-written business logic, and no injection/auth/data-exposure surface is touched. One earlier-flagged candidate (Dataset.inputSchema/expectedOutputSchema narrowed from Any to Dict[str, Any]) was marked "not independently ruled out" as a duplicate; on inspection the same narrowing was applied consistently to both the response (Dataset) and request (CreateDatasetRequest) models, which is consistent with a deliberate upstream spec change rather than an isolated bug, but I could not verify against the actual upstream spec. Given the large size (68 files, ~3300 insertions) and inability to independently confirm the regenerated types are fully faithful to the spec bump, a human look is still worthwhile even though no concrete bug was found.