Parallel Review Agents and Clearer Confidence Labels

MergeWatch 0.6 makes the review loop faster and the output more precise.

Added

• Security, bug, style, summary, and architecture agents now run concurrently.
• Findings include explicit confidence and severity fields.
• The final comment updates in place instead of creating a new comment on every run.
• Custom agents can read repository instructions from AGENTS.md.

Changed

Finding language now describes concrete evidence rather than assigning a blanket “safe” or “unsafe” status to the pull request. See why we made that change.

Upgrade notes

Self-hosted operators should pull the new image and run the included database migration before restarting workers. Existing .mergewatch.yml files remain compatible.