Verified all commands, file paths, class names, and Maven profiles
against the actual repository. Corrected the following:

- Thrift IDL is upstream (Hive TCLIService via kyuubi-relocated-thrift);
  rewrote the wire-format references accordingly. Only Kyuubi-side
  delta is python/scripts/thrift-patches/.
- SessionManager and OperationManager live in kyuubi-common, not
  kyuubi-server. Corrected high-sensitivity area paths.
- Replaced KyuubiBaseConnectionTest (does not exist) with the actual
  KyuubiStatementTest example.
- Replaced PlanOnlyStatementSuite (does not exist) with
  PlanOnlyOperationSuite.
- Replaced the broad "KYUUBI_UPDATE=1 build/mvn clean test" with
  dev/gen/gen_all_config_docs.sh, which is the scoped wrapper that
  actually regenerates settings.md.
- Rewrote the CI failure investigation section: Kyuubi CI does not
  emit per-test check-run annotations the way Spark CI does, so the
  recipe now uses gh run --log-failed and artifact download.
- Enumerated the externals/* engine modules explicitly instead of
  relying on a glob pattern that did not cover kyuubi-trino-engine.

Verified against root pom.xml, kyuubi-common/pom.xml,
ConfigBuilder.scala, and live source. Corrections:

- Removed externals/kyuubi-chat-engine reference; the CHAT engine was
  removed in Kyuubi 1.12 (per docs/deployment/migration-guide.md).
- Rephrased the engine profile matrix: only Spark and Scala have
  version profiles in pom.xml; Flink (1.20.3), Hive (3.1.3), and Trino
  client (411) are pinned to single versions.
- Java tests use JUnit 4 (no AssertJ dependency in any pom.xml);
  corrected the testing section accordingly.
- kyuubi-common does not contain in-tree Thrift IDL; described it as
  hosting relocated Hive Thrift RPC types instead.
- Distinguished integration-style tests from unit tests for build/dist
  prerequisite (per docs/contributing/code/testing.md).
- Server-only configs are declared via .serverOnly on the
  ConfigBuilder; serverOnlyConfEntries is the private registry that
  tracks them.
- Softened the "Never" rules for process.destroy() and
  SparkSession.active to "do not introduce new usage", since legacy
  usages exist in Spark plugin entry points that require a no-arg
  constructor.

Markdown reflows naturally on render; the hard wraps at ~70 chars
were a stylistic choice from the initial draft that made the source
harder to edit. Code fences, tables, nested list items, and the
license header are preserved unchanged.

- Server boundary: rephrase from "no Spark/Flink/Hive/Trino
  dependency" to "no Kyuubi engine module dependency". The server
  legitimately uses io.trino:trino-client for the Trino frontend and
  similar protocol/client libraries are allowed.
- Engine profile matrix: Flink also has version profiles
  (flink-1.17/1.18/1.19/1.20). Updated the table and the surrounding
  text; Hive and Trino remain pinned.
- Spark SQL syntax: Kyuubi extensions already add project-specific
  clauses (OPTIMIZE ... ZORDER BY). Rephrased from "no SQL syntax that
  Spark doesn't support" to require maintainer agreement and
  version-gated parser support for new clauses.
- Thrift dependencies: kyuubi-common pulls both
  kyuubi-relocated-thrift and kyuubi-relocated-hive-service-rpc;
  spelled both out.
- Utils.isTesting: legacy usages remain in production
  (EngineRef, JdbcProcessBuilder, TrinoProcessBuilder, BatchesResource).
  Softened from "never" to "do not introduce new ones without
  reviewer agreement".
- git config user.email: removed the "reviewers will block" framing,
  which is not documented as a hard policy; kept the practical fix-it
  guidance.

…d process.destroy primary rules

- Pre-flight: warn that Apache RAT scans the working tree and only
  honors .rat-excludes (not .gitignore), so private ignored files
  outside the repo root are safer.
- Scala specifics: rephrase the SparkSession.active rule to match the
  softened Never-list wording (allow Spark plugin/catalog entry
  points with no parameterized path).
- Error handling: rephrase the process.destroy() rule to "do not
  introduce direct process.destroy() / destroyForcibly()" for
  consistency with the Never list.

….md Never list

… Spark/Iceberg style

…NT>:<MODEL_ID>

- AGENTS.md: pin the example to the canonical model id (`claude-opus-4-7`).
- PULL_REQUEST_TEMPLATE: rename the field from `Generated-by:` to
  `Assisted-by:`, broaden the prompt to cover review/exploration, and
  align the example.