feat(executor): add support for Codex image generation tool usage tracking

- Introduced `publishCodexImageToolUsage` to report image generation tool metrics.
- Updated executor logic to handle image generation tool events and defaults.
- Added parsing logic for `image_gen` tool usage details in `helps/usage_helpers.go`.
- Updated `UsageReporter` for additional model-specific usage publishing.
- Refactored usage detail normalizations.

Closes: #3063
This commit is contained in:
Luis Pater
2026-04-26 22:19:03 +08:00
parent 38573050aa
commit c7b28ba058
2 changed files with 70 additions and 17 deletions

View File

@@ -30,8 +30,9 @@ import (
)
const (
codexUserAgent = "codex-tui/0.118.0 (Mac OS 26.3.1; arm64) iTerm.app/3.6.9 (codex-tui; 0.118.0)"
codexOriginator = "codex-tui"
codexUserAgent = "codex-tui/0.118.0 (Mac OS 26.3.1; arm64) iTerm.app/3.6.9 (codex-tui; 0.118.0)"
codexOriginator = "codex-tui"
codexDefaultImageToolModel = "gpt-image-2"
)
var dataTag = []byte("data:")
@@ -263,6 +264,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
if detail, ok := helps.ParseCodexUsage(eventData); ok {
reporter.Publish(ctx, detail)
}
publishCodexImageToolUsage(ctx, reporter, body, eventData)
completedData := eventData
outputResult := gjson.GetBytes(completedData, "response.output")
@@ -496,6 +498,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
if detail, ok := helps.ParseCodexUsage(data); ok {
reporter.Publish(ctx, detail)
}
publishCodexImageToolUsage(ctx, reporter, body, data)
data = patchCodexCompletedOutput(data, outputItemsByIndex, outputItemsFallback)
translatedLine = append([]byte("data: "), data...)
}
@@ -859,6 +862,31 @@ func ensureImageGenerationTool(body []byte, baseModel string, auth *cliproxyauth
return body
}
func publishCodexImageToolUsage(ctx context.Context, reporter *helps.UsageReporter, body []byte, completedData []byte) {
detail, ok := helps.ParseCodexImageToolUsage(completedData)
if !ok {
return
}
reporter.EnsurePublished(ctx)
reporter.PublishAdditionalModel(ctx, codexImageGenerationToolModel(body), detail)
}
func codexImageGenerationToolModel(body []byte) string {
tools := gjson.GetBytes(body, "tools")
if tools.IsArray() {
for _, tool := range tools.Array() {
if tool.Get("type").String() != "image_generation" {
continue
}
if model := strings.TrimSpace(tool.Get("model").String()); model != "" {
return model
}
break
}
}
return codexDefaultImageToolModel
}
func isCodexModelCapacityError(errorBody []byte) bool {
if len(errorBody) == 0 {
return false

View File

@@ -48,6 +48,18 @@ func (r *UsageReporter) Publish(ctx context.Context, detail usage.Detail) {
r.publishWithOutcome(ctx, detail, false)
}
func (r *UsageReporter) PublishAdditionalModel(ctx context.Context, model string, detail usage.Detail) {
if r == nil {
return
}
model = strings.TrimSpace(model)
if model == "" {
return
}
detail = normalizeUsageDetailTotal(detail)
usage.PublishRecord(ctx, r.buildRecordForModel(model, detail, false))
}
func (r *UsageReporter) PublishFailure(ctx context.Context) {
r.publishWithOutcome(ctx, usage.Detail{}, true)
}
@@ -65,15 +77,20 @@ func (r *UsageReporter) publishWithOutcome(ctx context.Context, detail usage.Det
if r == nil {
return
}
detail = normalizeUsageDetailTotal(detail)
r.once.Do(func() {
usage.PublishRecord(ctx, r.buildRecord(detail, failed))
})
}
func normalizeUsageDetailTotal(detail usage.Detail) usage.Detail {
if detail.TotalTokens == 0 {
total := detail.InputTokens + detail.OutputTokens + detail.ReasoningTokens
if total > 0 {
detail.TotalTokens = total
}
}
r.once.Do(func() {
usage.PublishRecord(ctx, r.buildRecord(detail, failed))
})
return detail
}
// ensurePublished guarantees that a usage record is emitted exactly once.
@@ -93,9 +110,16 @@ func (r *UsageReporter) buildRecord(detail usage.Detail, failed bool) usage.Reco
if r == nil {
return usage.Record{Detail: detail, Failed: failed}
}
return r.buildRecordForModel(r.model, detail, failed)
}
func (r *UsageReporter) buildRecordForModel(model string, detail usage.Detail, failed bool) usage.Record {
if r == nil {
return usage.Record{Model: model, Detail: detail, Failed: failed}
}
return usage.Record{
Provider: r.provider,
Model: r.model,
Model: model,
Source: r.source,
APIKey: r.apiKey,
AuthID: r.authID,
@@ -201,18 +225,15 @@ func ParseCodexUsage(data []byte) (usage.Detail, bool) {
if !usageNode.Exists() {
return usage.Detail{}, false
}
detail := usage.Detail{
InputTokens: usageNode.Get("input_tokens").Int(),
OutputTokens: usageNode.Get("output_tokens").Int(),
TotalTokens: usageNode.Get("total_tokens").Int(),
return parseOpenAIStyleUsageNode(usageNode), true
}
func ParseCodexImageToolUsage(data []byte) (usage.Detail, bool) {
usageNode := gjson.ParseBytes(data).Get("response.tool_usage.image_gen")
if !usageNode.Exists() || !usageNode.IsObject() {
return usage.Detail{}, false
}
if cached := usageNode.Get("input_tokens_details.cached_tokens"); cached.Exists() {
detail.CachedTokens = cached.Int()
}
if reasoning := usageNode.Get("output_tokens_details.reasoning_tokens"); reasoning.Exists() {
detail.ReasoningTokens = reasoning.Int()
}
return detail, true
return parseOpenAIStyleUsageNode(usageNode), true
}
func ParseOpenAIUsage(data []byte) usage.Detail {
@@ -220,6 +241,10 @@ func ParseOpenAIUsage(data []byte) usage.Detail {
if !usageNode.Exists() {
return usage.Detail{}
}
return parseOpenAIStyleUsageNode(usageNode)
}
func parseOpenAIStyleUsageNode(usageNode gjson.Result) usage.Detail {
inputNode := usageNode.Get("prompt_tokens")
if !inputNode.Exists() {
inputNode = usageNode.Get("input_tokens")