Escalate agent task failures from Haiku to Sonnet to Opus with tier logging.
npx skills add https://github.com/monktui/git-review --skill escalation-monktui
Please help me install this Agent Skill. Skill: escalation Source: https://github.com/monktui/git-review/tree/main/.claude/skills/escalation Command: npx skills add https://github.com/monktui/git-review --skill escalation-monktui
Confidence-based escalation prevents an AI agent from repeatedly failing at the same task level by moving to a more capable tier when clear failure signals appear.
In your agent run, follow the escalation signals to move the task from Haiku to Sonnet or Opus after a failed attempt that matches one of the defined criteria, while logging the reason for future routing improvements.
High-intent search queries and answers about installing and using this skill.
Agent escalation moves tasks from Haiku to Sonnet to Opus after explicit failure signals like test failures, schema mismatches, or repeated tool rejections. It enforces a one-attempt-per-tier rule to prevent skipping directly to the highest tier.
Failure signals that trigger escalation include test failures, irreconcilable lint warnings, wrong output schemas, repeated tech-lead blocks, timeouts, and repeated enforcement-hook blocks. These specific issues indicate the current model tier lacks capability to resolve the task.
Avoid escalation for size-only problems, missing context where adding instructions suffices, and infrastructure errors requiring a retry. Escalation targets model capability failures, not environmental or context-related issues.
Log agent escalation decisions by recording structured routing data to claude-mem. This tracks failure signals and tier transitions across multi-attempt workflows, providing diagnostic history to optimize future model selection.
You cannot skip tiers during agent escalation. The protocol enforces a strict Haiku to Sonnet to Opus progression with a one-attempt-per-tier rule, unless the task is explicitly known beforehand to require Opus capability.
Precommit gating works with model escalation by acting as an enforcement hook. When precommit gates fail repeatedly due to code execution issues or lint warnings, it triggers the escalation protocol to route the task to a higher model tier.
eysenfalkEscalate agent tasks from Haiku to Sonnet to Opus on failure signals.
ryanczakDecide task escalation actions from failure outcomes in milestone-based projects.
0r0looDiagnose AI agent failures and apply targeted recovery solutions.
ryasrkAssign LLM model tiers to agents based on task complexity and reasoning needs.
