Claude Skills by FgForrest

Use when there are open Dependabot "bump" PRs (maven and/or github_actions, authored by app/dependabot) to assess, test locally, and consolidate into one commit on dev, or when reviewing/resolving Dependabot security alerts. Triggers include "consolidate the bump PRs", "handle the dependabot PRs", "bump dependencies", "close the version-bump PRs". Leaves feature/fix PRs and frozen research artifacts untouched.

Use when adding a new field, enum, or configuration option to any evitaDB schema type (ReferenceSchema, EntitySchema, AttributeSchema, etc.). Provides a comprehensive 8-layer recipe covering contracts, DTOs, builders, mutations, external APIs (gRPC/GraphQL/REST) with backward compatibility, Kryo serializers, and WAL serializers.

Analyze GraphQL breaking changes between the public evitaDB demo server schema and the current local branch's schema. Builds the project, boots the local server, runs graphql-inspector via `tools/diff-graphql-schemas.sh`, and writes a prose summary of breaking changes and their client-facing implications. If the user provides a path to a downstream project that generates a GraphQL client from evitaDB's schema, the summary is tailored to that project specifically (TypeScript-heavy by default)....

Audit Kryo persistence backward compatibility before releasing a SNAPSHOT. Finds every class serialized through SerialVersionBasedSerializer whose on-disk format changed since the previous released version, and verifies each has its serialVersionUID bumped exactly once with a matching backward-compatible (BWC) reader registered in every Kryo configurer that registers it. Invoke when preparing a release, after adding/removing a persisted field, or when asked to check that stored catalogs / WAL...

Plans and executes implementation of a new evitaDB query constraint (filter, order, or require). Produces a task checklist covering all layers — query model, EvitaQL grammar, engine translator, Kryo serializer, external APIs, tests, and documentation — with exact file paths. Use when adding new filter/order/require constraints or variants of existing ones.

Use when adding a new response object (DTO surface) to evitaDB's external APIs — GraphQL, REST, and gRPC. Covers descriptors and their inheritance, interfaces and field-override rules, per-parameter object caches, GraphQL data fetchers, and gRPC proto/converter patterns with backward-compat rules. Use this whenever introducing a new object type, a suffix variant, or a polymorphic interface that must be exposed via at least two of the three APIs.

Analyze OpenAPI (REST) breaking changes between the public evitaDB demo server schema and the current local branch's schema. Builds the project, boots the local server, runs openapi-diff via `tools/diff-openapi-schemas.sh`, and writes a prose summary of breaking changes and their client-facing implications. If the user provides a path to a downstream project that generates a REST client from evitaDB's OpenAPI schema, the summary is tailored to that project specifically (TypeScript-heavy by de...

Generate enriched release notes for a major or patch release of evitaDB. Runs the deterministic skeleton generator and adds short user-facing prose to all breaking changes and to the features deemed "interesting", then writes the result to release-notes.md. Invokable locally as /release-notes and headlessly from the CI release workflow.

Prepare or update a release PR from dev to master with auto-generated release notes

Replay upstream RoaringBitmap changes onto the vendored evita_roaring_bitmap module. Walks upstream/master since the recorded base, triages each commit against the vendored subset, incorporates relevant fixes, and updates the sync ledger. Invoke when checking for or applying upstream RoaringBitmap updates.

Measure evitaDB write-path (commit/transaction) performance by replaying a real production catalog's Write-Ahead Log against an embedded instance via the WalReplayBenchmark JMH harness, with async-profiler (alloc/cpu) or JMH `-prof gc`. Invoke when profiling commit latency, allocation, or GC behavior on the write path, when a production WAL export needs re-measuring, or when a profiling run produced a suspicious/empty/inconsistent result (zero-byte collapsed dump, contended-box numbers, cross...

Measure how long a full catalog reindex-and-publish takes — re-upsert every entity of a real production catalog into a fresh catalog in WARM_UP mode through the gRPC driver, then transition it to ALIVE via goLive. Reports load wall-clock, per-collection and per-upsert latency distributions, and the goLive transition separately. Invoke when asked how long publishing a dataset takes, when verifying a full reindex still completes after write-path or driver changes, when comparing ingestion throu...