Compare SeaTunnel Plugin Baseline

Use this skill when the user wants to compare a new official SeaTunnel version with the local plugin baseline template, detect connector and dependency changes, and generate a human-reviewable update proposal.

Goal

Produce a review-first diff for:

• connector catalog changes
• plugin-mapping.properties changes
• connector POM dependency changes (only treat provided as automatic diff input)
• JDBC profile changes from connector-jdbc/pom.xml and dialect source additions
• candidate updates to local seed JSON / baseline rules

Do not silently overwrite the local baseline. Always stop for human review before applying.

Primary sources

Only use official sources:

• apache/seatunnel Git tags
• apache/seatunnel-website versioned docs (when docs help explain dependencies)

Prefer tag source files over rendered docs.

Inputs to confirm

Ask or infer:

• target version (required)
• baseline version (default: latest value in reviewed_versions)
• scope:
  • full connector catalog
  • JDBC profiles only
  • dependency diff only
  • all of the above

Local files to inspect first

Read only what is needed:

• internal/apps/plugin/seed/seatunnel-plugins.json
• .trellis/tasks/03-16-plugin-official-dependency-baseline/design.md
• related plugin service code if target-dir or baseline rules matter

Comparison workflow

1. Fetch connector directories for both versions from official tag source.
   • Compare seatunnel-connectors-v2/**/pom.xml module names
   • Compare plugin-mapping.properties
2. Build connector diff sets:
   • added
   • removed
   • renamed / remapped
   • unchanged
3. For each changed/new connector, inspect the module POM:
   • dependencies excluding test
   • only elevate provided dependencies to automatic review candidates
   • treat compile/runtime deps as auxiliary signals only, not as direct baseline-update evidence
4. For JDBC:
   • diff seatunnel-connectors-v2/connector-jdbc/pom.xml
   • inspect dialect factory additions/removals in source
   • map database profile -> artifacts -> version range
5. Classify each result into one of:
   • no extra dependency
   • companion connector only
   • isolated dependency under plugins/<mapping>
   • shared dependency under lib/
   • manual review required
6. Apply existing project rules while classifying:
   • built-in package jars are excluded from extra dependency recommendations
   • file-base / file-base-hadoop are hidden companion connectors
   • http-base remains visible but acts as companion for http-*
   • do not auto-promote ambiguous JDBC cases to defaults
   • POM compile/runtime dependency changes do not auto-generate baseline changes
7. Generate a human review table before any patch.

Required output format

Always output these sections:

1. Summary

• target version
• baseline version
• connector diff counts
• dependency diff counts

2. Connector diff table

Columns:

• connector
• change type
• mapping change
• baseline impact
• review needed

3. Dependency/profile diff table

Columns:

• connector/profile
• artifact(s)
• old rule
• new official signal (provided, JDBC source change, or manual note)
• recommended action

4. Proposed JSON patch plan

Group into:

• add connector/profile
• remove connector/profile
• update versions
• update target_dir policy
• mark manual-review

5. Human confirmation questions

Ask concise batch questions such as:

• approve all exact version bumps?
• keep ambiguous connectors as manual-review?
• apply JDBC profile additions now?

When the user approves changes

Only then:

• update the seed JSON / baseline files
• update design docs if rules changed
• keep the patch scoped to the approved rows

Guardrails

• Do not treat local baseline as truth for connector catalog freshness.
• Do not infer external dependency delivery from docs alone if tag source contradicts it.
• Do not auto-apply new JDBC profiles without showing the table first.
• Do not mix “official signal” and “project override” in the same column.
• Do not promote compile/runtime dependency changes from POM into automatic baseline updates.
• If evidence is weak, classify as manual_review.