From 45676fdc8da710b7c046cf4669ad1f4be5cb4f59 Mon Sep 17 00:00:00 2001 From: shaw Date: Mon, 26 Jan 2026 11:04:38 +0800 Subject: [PATCH 1/2] =?UTF-8?q?fix(ci):=20=E8=BD=AC=E4=B9=89=20Telegram=20?= =?UTF-8?q?=E6=B6=88=E6=81=AF=E4=B8=AD=E7=9A=84=20Markdown=20=E7=89=B9?= =?UTF-8?q?=E6=AE=8A=E5=AD=97=E7=AC=A6?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 修复发布通知发送失败的问题，原因是 tag message 中包含未闭合的 Markdown 格式标记（如 user_id 中的 _ 被解析为斜体开始）导致 Telegram API 返回解析错误。 添加 sed 命令转义 _、*、` 和 [ 字符，避免被 Telegram Markdown 解析器错误处理。 --- .github/workflows/release.yml | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml index 73ca35d9..0415000d 100644 --- a/.github/workflows/release.yml +++ b/.github/workflows/release.yml @@ -222,8 +222,9 @@ jobs: REPO="${{ github.repository }}" GHCR_IMAGE="ghcr.io/${REPO,,}" # ${,,} converts to lowercase - # 获取 tag message 内容 + # 获取 tag message 内容并转义 Markdown 特殊字符 TAG_MESSAGE='${{ steps.tag_message.outputs.message }}' + TAG_MESSAGE=$(echo "$TAG_MESSAGE" | sed 's/\([_*`\[]\)/\\\1/g') # 限制消息长度（Telegram 消息限制 4096 字符，预留空间给头尾固定内容） if [ ${#TAG_MESSAGE} -gt 3500 ]; then From 56a1e29cdddd21f22b0923679fc06ce350dcd602 Mon Sep 17 00:00:00 2001 From: shaw Date: Tue, 27 Jan 2026 09:16:34 +0800 Subject: [PATCH 2/2] =?UTF-8?q?fix(gateway):=20=E4=BF=AE=E5=A4=8D=20SSE=20?= =?UTF-8?q?=E6=B5=81=E5=BC=8F=E5=93=8D=E5=BA=94=20usage=20=E7=BB=9F?= =?UTF-8?q?=E8=AE=A1=E9=94=99=E8=AF=AF?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit message_delta 应完全覆盖 message_start 的 usage 数据， 而非仅在值为 0 时才更新。 --- backend/internal/service/gateway_service.go | 17 +++++------------ 1 file changed, 5 insertions(+), 12 deletions(-) diff --git a/backend/internal/service/gateway_service.go b/backend/internal/service/gateway_service.go index a8f5baeb..5819f15a 100644 --- a/backend/internal/service/gateway_service.go +++ b/backend/internal/service/gateway_service.go @@ -3372,19 +3372,12 @@ func (s *GatewayService) parseSSEUsage(data string, usage *ClaudeUsage) { } `json:"usage"` } if json.Unmarshal([]byte(data), &msgDelta) == nil && msgDelta.Type == "message_delta" { - // output_tokens 总是从 message_delta 获取 + // message_delta 是推理结束后的最终统计，应完全覆盖 message_start 的数据 + // 这对于 Claude API 和 GLM 等兼容 API 都是正确的行为 + usage.InputTokens = msgDelta.Usage.InputTokens usage.OutputTokens = msgDelta.Usage.OutputTokens - - // 如果 message_start 中没有值，则从 message_delta 获取（兼容GLM等API） - if usage.InputTokens == 0 { - usage.InputTokens = msgDelta.Usage.InputTokens - } - if usage.CacheCreationInputTokens == 0 { - usage.CacheCreationInputTokens = msgDelta.Usage.CacheCreationInputTokens - } - if usage.CacheReadInputTokens == 0 { - usage.CacheReadInputTokens = msgDelta.Usage.CacheReadInputTokens - } + usage.CacheCreationInputTokens = msgDelta.Usage.CacheCreationInputTokens + usage.CacheReadInputTokens = msgDelta.Usage.CacheReadInputTokens } }