Cyberclaw — Case Study
Scroll to explore ↓
Agent knows to SSH into relevant dev VM, and reads the relevant SoP restriction that editing e2e user story tests is strictly prohibited
Agent knows to commit current repo state if positive progress made — else, it leaves the repo state uncommitted, just 'resetting the deployed infra state for a fresh run' and then spawning another coding agent through automate_loop.sh
Calling automate_loop.sh with relevant paths + a succinct summary of the prior agent's goal
Orchestrator behavior: if prior coding CLI succeeded, runs solution.sh to determine current progress and saves result in error_output location; if prior coding CLI failed, skips validation as spawned agent will handle it
Agent delivers complete success without human intervention
Agent knows to SSH into relevant dev VM, and reads the relevant SoP restriction that editing e2e user story tests is strictly prohibited