HypAware
View as Markdown

Review an agent run

When an agent surprises you, good or bad, find out why.

Each recorded run shows why the agent did what it did: the prompt it had, what it tried, in what order, and where the time went. People read one run closely to improve their prompts, skills, and harnesses.

Ask your agent

From my HypAware history, walk me through [job]'s last run. What did the agent try, in what order, and where did the time go?

From my HypAware history, how did the agent decide to [change] in [topic]? What did it see that led there?

From my HypAware history, in [session], what explanations did the agent consider for [problem], which did it rule out, and was it right to?

From a terminal:

Terminal
hyp ask "From my HypAware history, walk me through [job]'s last run. What did the agent try, in what order, and where did the time go?"

What you get back

A step-by-step account of one run from its recorded messages and tool calls: what the agent was asked, what it read, what it tried, what failed, and how long each stretch took. Ask what it would change in the prompt or harness.

Get a better answer

  • Point at one run: a session id, a time, or the job name and "last run."
  • Subagent work is recorded with the session. Ask the agent to include it.
  • For the same failure across many runs, ask about the pattern instead: "From my HypAware history, where do my agents keep getting stuck?"
Was this page useful?

↑↓ navigate↵ openesc close