model-escalation

name	model-escalation
description	Advisory model escalation based on task outcome tracking
scope	core
user-invocable	false

Model Escalation Skill

Tracks task outcomes and advises reasoning-effort changes when failures are detected. Advisory-only — the orchestrator makes the final decision (R010), and OMX remains the owner of concrete model resolution.

Escalation Path

none → minimal → low → medium → high → xhigh → ultra → max

Trigger Conditions

Condition	Action
2+ failures with same effort for same agent type	Advise escalation
3+ consecutive failures across any agent type	Advise global escalation
Sustained success after escalation	Advise de-escalation

Thresholds

Parameter	Default	Description
failure_threshold	2	Failures before escalation advisory
consecutive_threshold	3	Consecutive failures for global advisory
cooldown_tasks	5	Successes before de-escalation advisory

Cost Guard

• Advisory identifies the next supported reasoning effort
• De-escalation suggested after sustained success at higher tier
• Outcomes are tracked per session via a PPID-scoped temp file

Architecture

PostToolUse (Task) → task-outcome-recorder.sh
  Records: agent_type, model_reasoning_effort, success/failure, error_summary
  Storage: /tmp/.codex-task-outcomes-$PPID (JSON lines, max 50)

PreToolUse (Task) → model-escalation-advisor.sh
  Reads outcomes → counts failures → advises escalation via stderr
  Advisory only — never blocks, never modifies tool input

Advisory Format

--- [Model Escalation Advisory] ---
  Agent type: {agent_type}
  Current effort: {current_effort}
  Recent failures: {count}/{threshold}
  Recommended effort: {next_effort}
  Concrete model: resolved by the active OMX runtime
---

The advisor must not rewrite a concrete model ID. If the active model does not support the recommended value, the orchestrator chooses the nearest supported effort from the generated runtime model table.