Local review
Review working-tree changes, explain risks, point to exact lines, and avoid low-confidence noise.
On the roadmap
Debug mode already focuses Agent on investigation. Dedicated review workflows still need stronger evidence collection, severity handling, and focused user interfaces beyond the generic chat surface.
Review working-tree changes, explain risks, point to exact lines, and avoid low-confidence noise.
Instrument an app, reproduce behavior, gather runtime evidence, explain the cause, and remove temporary instrumentation.
Agent, Tab, Quick Edit, model choice, MCP tools, and checkpoints are available now.