new-plugin

name: new-plugin description: Add a new input, event, or output plugin to Eventum -- orchestrate agents through research, design, code, UI, review, docs. user-invokable: true argument-hint: "/ (e.g. input/kafka, output/s3, event/jsonl)"

Add New Plugin

Orchestrate the creation of a new Eventum plugin $ARGUMENTS by delegating to your team of agents.

Parse the argument as <type>/<name> where type is input, event, or output.

Phase 1: Research

Delegate to researcher agent:

• Study existing plugins of the same type: read base classes, config patterns, test structure
• Research the external technology (library, protocol, API) the plugin will integrate with
• Identify dependencies, authentication patterns, configuration options
• Report findings with recommendations

Phase 2: Design

Delegate to architect agent:

• Design the plugin based on researcher's findings: config model, implementation approach, test strategy
• Follow plugin conventions: Pydantic config (frozen, extra='forbid'), class extending base plugin
• Plan the UI integration (Zod schema, form component)
• Produce implementation plan with ordered steps

TL directly: Present the design to the user for approval. Ask whether to implement UI changes or skip for now.

Phase 3: Implement

Delegate to developer agent:

• Create plugin directory: eventum/plugins/<type>/plugins/<name>/
  • config.py -- Pydantic config model
  • plugin.py -- Plugin class extending the base
  • __init__.py -- Exports
  • tests/ -- Test directory structure
• If UI approved:
  • Zod schema mirroring the Pydantic config
  • UI form component
  • Registry entries (union index, default config, plugin info)
• Run ruff/mypy on own code before returning

Checkpoint: Present the developer's changes to the user before proceeding.

Phase 4: Test

Delegate to qa-engineer agent:

• Write comprehensive tests in eventum/plugins/<type>/plugins/<name>/tests/
• Test plugin functionality, config validation, edge cases, error paths
• Run: uv run pytest eventum/plugins/<type>/plugins/<name>/tests/ -v
• Run: uv run ruff check and uv run mypy on all changed files
• Report results

If QA reports failures: route findings to developer to fix, then re-run QA. Loop until all checks pass. If the loop does not converge after 3 cycles, stop and consult the user.

Phase 5: Code Review

Delegate to code-reviewer agent:

• Review ALL changes: Python plugin code + UI code (if any) + tests
• If verdict is FAIL: route findings to developer and/or qa-engineer to fix, then re-review
• Loop until PASS. If the loop does not converge after 3 cycles, stop and consult the user.

This is a mandatory quality gate -- do NOT skip it.

Phase 6: Documentation

Delegate to docs-writer agent:

• Create MDX page at ../docs/content/docs/plugins/<type>/<name>.mdx:
  • Overview, configuration table, examples (basic + advanced), input/output example
• Add entry to ../docs/content/docs/plugins/<type>/meta.json
• Add changelog entry to CHANGELOG.md under ## Unreleased / ### New Features

Phase 7: Update CLAUDE.md

Delegate to developer agent:

• Update CLAUDE.md (this repo): plugin tables, counts, new dependency if applicable

Phase 8: Final Verification

Delegate to qa-engineer agent:

• Run full pipeline:

  uv run pytest eventum/plugins/<type>/plugins/<name>/tests/ -v
  uv run ruff check eventum/plugins/<type>/plugins/<name>/
  uv run mypy eventum/plugins/<type>/plugins/<name>/
  cd ../docs && pnpm build
  

• Report all-green status

If any check fails: route to the responsible agent (developer for code, docs-writer for docs), fix, and re-verify. If the loop does not converge after 3 cycles, stop and consult the user.

Phase 9: Summary

TL directly:

Present to the user:

• What was created (plugin code, UI, tests, docs)
• File list with key changes
• Verification results

Important

• Do NOT commit or push unless the user explicitly asks.
• Track progress with the todo list throughout.
• If blocked or uncertain, ask the user rather than guessing.
• Check CLAUDE.md cross-cutting change checklist for completeness.