Save Usage ~4x (Codex oAuth users)

Classifies tasks by safety and complexity to route safe, negligible queries to gpt-5.1-codex-mini and escalates critical or executable tasks to gpt-5.3-codex.

This skill is coherent and does what it says: it classifies inputs and forwards risky/executable/high-impact tasks to a stronger model. Before installing, confirm: (1) your platform actually provides the sessions_spawn() API and you understand what it will send (it forwards the full task prompt to gpt-5.3-codex); (2) you're comfortable with potential privacy and billing implications of more frequent escalations; (3) you want the agent to make automatic escalation decisions rather than prompting you first. Test with non-sensitive prompts to see how often it escalates and monitor costs and logs. If you need stricter privacy, require explicit user confirmation before escalation or redact sensitive material before forwarding.

Type: OpenClaw Skill Name: save-usage Version: 1.0.0 The skill bundle defines a model routing strategy for an AI agent, instructing it to escalate sensitive or high-impact tasks (e.g., code execution, handling secrets, security, architecture) to a more capable model (`gpt-5.3-codex`) via `sessions_spawn()`. Conversely, it restricts a less capable model (`gpt-5.1-codex-mini`) to only safe, negligible, and non-executable tasks. The instructions in SKILL.md are designed to enhance security and resource management by preventing the agent from performing risky actions on an inappropriate model. There is no evidence of malicious intent, data exfiltration, unauthorized execution, or harmful prompt injection.