Add Visual Companion as a visual decision I/O plugin

[Mineru98]

Add Visual Companion as a visual decision I/O plugin

Fixes #130

Summary

Adds visual-companion, an Ouroboros-native plugin for browser-backed UI/UX decision prompts.

The plugin lets a workflow start a local browser session, publish an agent-authored HTML screen, wait for a user click/form event, recover that answer as JSON, and attach handoff/audit evidence for downstream agents.

Problem

Text-only UI/UX questions are often ambiguous for people who do not already share frontend vocabulary with the agent. Terms like layout density, interaction state, selection behavior, or visual hierarchy can sound acceptable in prose while behaving differently once implemented.

This is especially common for non-designers, frontend beginners, and vibe coders. The wrong decision often becomes visible only after code has already been written.

Changes

• Add plugins/visual-companion with a valid ouroboros.plugin.json.
• Copy the existing visual-companion skill assets into the plugin package.
• Add a Python adapter exposing:
  • start
  • show
  • wait
  • read
  • stop
• Persist session state and handoff artifacts for completed, blocked, and failed command paths.
• Keep the private browser event token out of public stdout and handoff payloads.
• Register the plugin in catalog/index.json.
• Add focused tests for manifest shape, CLI parsing, blocked handoff behavior, fresh screen names, and a real Node-backed browser-event lifecycle.

Validation

py -3 -m unittest tests.test_visual_companion_plugin
py -3 scripts\validate_contract.py
git diff --check
py -3 -m py_compile plugins\visual-companion\visual_companion_plugin\__main__.py plugins\visual-companion\visual_companion_plugin\__init__.py

Also smoke-tested isolated plugin installation with a temporary plugin home:

ouroboros plugin add . --plugin visual-companion --plugin-home-root <temp> --trust-root <temp> --lockfile <temp> --catalog-state <temp>

Notes

The local python/pip shim in this environment points at a broken E:\SourceCode\ai-workflow\.venv, so validation used the Windows py -3 launcher.