escalation

Model tier escalation for AI workflows progresses tasks from Haiku to Sonnet to Opus based on failure signals. It enforces a never-skip-tiers rule, escalating sequentially only when specific triggers like test failures, timeouts, or repeated hook blocks occur.

Escalation triggers include test failures, timeouts, and repeated hook blocks. When a lower-tier agent encounters these failure signals, the task is escalated to the next model tier along with context on the previous failure.

Use the escalation protocol to re-assign a task to the next model tier when the current agent fails. When a junior coder agent fails a bug fix after one retry, the task automatically escalates to a more capable coder agent with failure context.

No, the protocol enforces a never-skip-tiers rule requiring sequential progression from Haiku to Sonnet to Opus. This prevents direct jumps to higher tiers, ensuring structured failure recovery and efficient resource utilization.

Yes, a diagnostic skill is required for failure analysis during model tier escalation. It works alongside the escalation protocol to analyze failure signals and logs all escalations for pattern review.

The escalation protocol logs all tier escalations for pattern review. By recording when tasks progress from Haiku to Sonnet to Opus based on triggers, it provides structured data to analyze agent performance and recurring failure signals.