Agent Loop
How YuvaDev agents execute work
The execution loop keeps planning, implementation, and verification tightly connected.
Execution diagram
Prompt
→
Plan
→
Edits + Verify
Feedback loop when checks fail
Core loop stages
- 1. Parse the intent and gather context from your workspace.
- 2. Produce a concrete plan with explicit validation steps.
- 3. Apply targeted code edits while preserving project conventions.
- 4. Run build/tests/type checks and capture failures.
- 5. Iterate until acceptance criteria pass or user intervention is required.
Pseudo-code
while (task.notDone) { plan = agent.proposePlan(context) approval = user.review(plan) if (!approval.ok) { continue } edits = agent.applyCodeChanges(plan) results = agent.runVerification(edits) if (results.pass) { task.complete() } else { context = agent.learnFrom(results) }}