Docs/AI Agent

AI Agent

CODVIAN turns an instruction into a visible sequence of project-aware actions, with review points where control matters.

Understand

The agent reads project files, instructions, and recent context before proposing a path forward. It can ask for clarification when the goal or constraints are ambiguous.

Plan

The work is broken into tasks that can be inspected and adjusted. A useful plan names the files, commands, expected output, and verification method.

  • Keep tasks small enough to review
  • State assumptions before making broad changes
  • Use project instructions when they exist
  • Ask before crossing a permission boundary

Build, run, and verify

CODVIAN can edit files, run terminal commands, inspect output, and use browser workflows where supported. Verification is part of the task, not an optional final step.

Human review remains important

Agent output should be reviewed like any other change. Inspect the diff, test the behavior, and approve the result before moving on.