jlcodes99/cockpit-tools v1.3.46 on GitHub

更新日志（中文）

默认使用 gpt-5.5 转发生图请求：由于官方 ChatGPT 账号通道已不再稳定支持 gpt-5.4-mini 作为这条生图链路的基础模型，原先转换后的请求可能因基础模型不兼容被上游拒绝；本地 API 服务现改用 gpt-5.5 作为 Responses 基础模型，同时保留 gpt-image-2 作为生图工具模型。
提升 Codex 协议兼容性：清理上游不接受的 JSON Schema 方言元字段，按 Codex 规则清洗工具名称，在唯一匹配时恢复工具命名空间，保留 service_tier 和缓存写入 token 用量，函数工具未提供 strict 时明确传递 false，并将提示可重试的 server error 纳入 Codex 账号故障切换。
增强 Codex 故障与协作处理：简化大型常量工具 Schema 联合结构，识别静默的零 token response.incomplete 终止，将配额故障限定到受影响的 credential，并在多 Agent 优化后恢复带点号的 collaboration 工具名称。

Route image-generation requests through gpt-5.5 by default: the official ChatGPT account channel no longer reliably accepts gpt-5.4-mini as the base model for this image path, so the previous transformed request could be rejected upstream with an incompatibility error. The local API Service now uses gpt-5.5 as the Responses base model while keeping gpt-image-2 as the image tool model.
Improve Codex protocol compatibility: normalize tool schemas by removing unsupported JSON Schema dialect metadata, sanitize tool names for Codex upstream rules, recover uniquely identifiable tool namespaces, preserve service_tier and cache-write token usage, explicitly forward omitted function-tool strict=false, and classify retry-advised server errors as eligible for Codex account failover.
Harden Codex failure and collaboration handling: simplify large constant tool-schema unions, detect silent zero-token response.incomplete terminations, keep quota failures scoped to the affected credential, and restore dotted collaboration tool names after multi-agent optimization.