agent-management

Prevent autonomous agents from shirking by cross-referencing ntfy notifications, agent_runner logs, and git history to detect premature completion or closure mechanisms. Apply structural fixes across all runs by patching prompts and playbooks, ensuring changes are grounded in transcript evidence to close gaps and prevent future shirking.

Worker agents produce trivial work or fail tasks due to structural causes like memory poisoning, premature completion, or closure mechanisms. Identify these behavioral failures by auditing prompts, playbooks, memories, and logs to fix the underlying structural issues rather than treating symptoms.

Audit agent logs and transcripts by running a workflow on a failing run to locate root causes, verifying findings against transcript evidence, and applying structural prompt or playbook fixes. Cross-reference ntfy notifications, agent_runner logs, and git history to ensure fixes are grounded in evidence and rolled back if they reintroduce failures.

Git history and ntfy notifications are essential for effective agent behavior auditing, as they allow you to monitor operational reliability, cross-reference agent_runner logs, and verify that structural prompt fixes are grounded in transcript evidence and rolled back if failures reappear.

The best way to patch agent playbooks and memories is to ground all changes in transcript evidence and research, apply fixes across all runs, and roll back immediately if the patched prompts or playbooks reintroduce previous behavioral failures or closure mechanisms.