Guide: repeated-task kickoff

This is the strongest first demo for Aionis.

Scenario

You have a recurring class of coding work:

• export bugs
• flaky tests
• billing retries
• migration regressions

You want the agent to stop rediscovering the same starting path every time.

Flow

1. write execution memory from a prior successful or partially successful run
2. optionally record tool feedback
3. call planningContext, experienceIntelligence, or taskStart for a similar task later
4. inspect whether the first action is stronger than a cold start

What success looks like

The agent returns:

• a tighter selected tool
• a better ordered candidate list
• a file-level next move
• a clearer rationale for why that move is grounded in prior execution

Best supporting evidence

The benchmark report is the best proof source for this flow:

• Memory reference
• Validation and Benchmarks

That report already calls out kickoff hit rate, path hit rate, stale-memory interference rate, and repeated-task cost reduction.