Free resources · For developers
Orchestrating subagents to audit an application
A reusable workflow for finding defects nobody reported: how it runs, what it costs, where it breaks.
The idea
One agent does the work. A separate subagent, in a fresh context, checks it against written criteria. The agent that did the work does not grade it.
The steps
- Write the criteria first.
- Split the application into parts.
- Give each part to its own subagent.
- Each subagent returns findings with evidence.
- A reviewer subagent checks each finding before it is reported.
What a finding holds
Where it is, why it matters, and how it was checked. The report also says what was not checked.
