Agent Debug Playbook

A debugging methodology for AI agents that are not behaving as expected. Covers 12 failure modes with a diagnostic process and resolution pattern for each.

Failure modes covered: tool call failures, context window exhaustion, memory loading issues, coordination loops between agents, authority scope violations, cron job silent death, and broken MCP server registration. Each has observable signals, a diagnostic prompt, and a specific fix pattern.

Also includes the self-diagnostic technique: instead of asking the agent to fix a problem, ask it to diagnose its own reasoning failure with no action mandate. This forces the agent to reason backward through its logic chain and surfaces instrument failures that standard debugging prompts miss.

What is included: the full playbook, the self-diagnostic prompt template, a pre-flight checklist for new agent deployments, and a session log format for capturing failure patterns.

Best for operators dealing with agents that produce wrong outputs, stop responding, or behave inconsistently across sessions.