Files
CLIProxyAPI/internal/runtime/executor/helps/antigravity_compaction.go
Luis Pater 344e0ef929 feat(executor): support Codex remote-compaction v2 on Kimi and Meta Responses routes
- Intercept compaction_trigger items on Kimi and Meta OpenAI Responses execution paths to synthesize sealed CPA compaction capsules and reject incomplete upstream summaries.
- Expand replayed CPA compaction capsules into developer context messages and drop foreign compaction items across Kimi and Meta execution and token-counting paths.

Closes: #6781
2026-10-19 22:07:14 +08:00

571 lines
22 KiB
Go

package helps
import (
"bytes"
"crypto/aes"
"crypto/cipher"
"crypto/rand"
"crypto/sha256"
"encoding/base64"
"encoding/json"
"fmt"
"io"
"strings"
"time"
cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v8/sdk/cliproxy/executor"
log "github.com/sirupsen/logrus"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
)
const (
antigravityCompactionCapsulePrefix = "cpa-ag-compact-v1:"
fixedCompactionKeySecret = "CLIProxyAPI"
)
type antigravityCompactionCapsuleData struct {
Summary string `json:"summary"`
Model string `json:"model"`
CreatedAt int64 `json:"created_at"`
}
// RecognizedAntigravityCompactionCapsule reports whether encrypted content uses the CPA capsule prefix.
func RecognizedAntigravityCompactionCapsule(encryptedContent string) bool {
return strings.HasPrefix(strings.TrimSpace(encryptedContent), antigravityCompactionCapsulePrefix)
}
// HasResponsesCompactionTrigger checks whether input contains a compaction_trigger item.
func HasResponsesCompactionTrigger(payload []byte) bool {
input := gjson.GetBytes(payload, "input")
if !input.IsArray() {
return false
}
for _, item := range input.Array() {
if item.Get("type").String() == "compaction_trigger" {
return true
}
}
return false
}
// HasResponsesCompactionItem checks whether input contains a compaction item.
func HasResponsesCompactionItem(payload []byte) bool {
input := gjson.GetBytes(payload, "input")
if !input.IsArray() {
return false
}
for _, item := range input.Array() {
if item.Get("type").String() == "compaction" {
return true
}
}
return false
}
// HasResponsesCompactionTriggerRequest reports whether req.Payload or opts.OriginalRequest contains a compaction_trigger item.
func HasResponsesCompactionTriggerRequest(req cliproxyexecutor.Request, opts cliproxyexecutor.Options) bool {
return HasResponsesCompactionTrigger(req.Payload) || HasResponsesCompactionTrigger(opts.OriginalRequest)
}
// ResponsesCompactionSourcePayload selects the source payload containing the compaction trigger or request history.
func ResponsesCompactionSourcePayload(req cliproxyexecutor.Request, opts cliproxyexecutor.Options) []byte {
payload := req.Payload
if !HasResponsesCompactionTrigger(payload) && HasResponsesCompactionTrigger(opts.OriginalRequest) {
payload = opts.OriginalRequest
}
if len(payload) == 0 && len(opts.OriginalRequest) > 0 {
payload = opts.OriginalRequest
}
return payload
}
// DropForeignResponsesCompactionItems removes compaction items whose encrypted_content does not use the CPA capsule prefix.
func DropForeignResponsesCompactionItems(payload []byte, provider string) []byte {
input := gjson.GetBytes(payload, "input")
if !input.IsArray() {
return payload
}
dropped := 0
items := make([]string, 0, len(input.Array()))
for _, item := range input.Array() {
if item.Get("type").String() == "compaction" && !RecognizedAntigravityCompactionCapsule(item.Get("encrypted_content").String()) {
dropped++
continue
}
items = append(items, item.Raw)
}
if dropped == 0 {
return payload
}
log.Warnf("%s compaction: dropped %d non-CPA compaction item(s)", provider, dropped)
out, _ := sjson.SetRawBytes(payload, "input", []byte("["+strings.Join(items, ",")+"]"))
return out
}
// ExpandResponsesCompactionCapsules drops foreign compaction items and expands recognized CPA compaction capsules in req and opts.
func ExpandResponsesCompactionCapsules(req *cliproxyexecutor.Request, opts *cliproxyexecutor.Options, provider string) error {
sameOriginal := req != nil && opts != nil && len(opts.OriginalRequest) > 0 && bytes.Equal(req.Payload, opts.OriginalRequest)
if req != nil {
req.Payload = DropForeignResponsesCompactionItems(req.Payload, provider)
if HasResponsesCompactionItem(req.Payload) {
expanded, errExpand := ExpandAntigravityCompactionCapsules(req.Payload)
if errExpand != nil {
return errExpand
}
req.Payload = expanded
}
}
if opts != nil && len(opts.OriginalRequest) > 0 {
if sameOriginal {
opts.OriginalRequest = req.Payload
return nil
}
opts.OriginalRequest = DropForeignResponsesCompactionItems(opts.OriginalRequest, provider)
if HasResponsesCompactionItem(opts.OriginalRequest) {
expanded, errExpand := ExpandAntigravityCompactionCapsules(opts.OriginalRequest)
if errExpand != nil {
return errExpand
}
opts.OriginalRequest = expanded
}
}
return nil
}
// PrepareAntigravityCompactionSummaryPayload prepares a payload for non-stream Antigravity summary generation.
func PrepareAntigravityCompactionSummaryPayload(payload []byte, modelName string) []byte {
out := payload
summaryPrompt := `{"type":"message","role":"user","content":[{"type":"input_text","text":"Please provide a concise and comprehensive summary of the preceding conversation and task progress so far, including user goals, key findings, actions taken, and current status, so that work can continue smoothly."}]}`
input := gjson.GetBytes(out, "input")
if input.IsArray() {
var filtered []string
for _, item := range input.Array() {
itemType := item.Get("type").String()
if itemType == "compaction_trigger" {
continue
}
filtered = append(filtered, item.Raw)
}
filtered = append(filtered, summaryPrompt)
newInput := "[" + strings.Join(filtered, ",") + "]"
out, _ = sjson.SetRawBytes(out, "input", []byte(newInput))
} else if input.Type == gjson.String && input.String() != "" {
userMsg := []byte(`{"type":"message","role":"user","content":[{"type":"input_text"}]}`)
userMsg, _ = sjson.SetBytes(userMsg, "content.0.text", input.String())
newInput := "[" + string(userMsg) + "," + summaryPrompt + "]"
out, _ = sjson.SetRawBytes(out, "input", []byte(newInput))
} else {
newInput := "[" + summaryPrompt + "]"
out, _ = sjson.SetRawBytes(out, "input", []byte(newInput))
}
// Remove fields not needed or incompatible with summary generation
for _, field := range []string{
"stream",
"tools",
"tool_choice",
"previous_response_id",
"parallel_tool_calls",
"additional_tools",
"truncation",
"metadata",
} {
out, _ = sjson.DeleteBytes(out, field)
}
out, _ = sjson.SetBytes(out, "stream", false)
return out
}
// deriveAntigravityCompactionKey derives an AES-256 key from the fixed CLIProxyAPI secret.
func deriveAntigravityCompactionKey() []byte {
h := sha256.Sum256([]byte(fixedCompactionKeySecret))
return h[:]
}
// SealAntigravityCompaction encrypts summary data into an opaque capsule using AES-GCM.
func SealAntigravityCompaction(summary, modelName string) (string, error) {
data := antigravityCompactionCapsuleData{
Summary: summary,
Model: modelName,
CreatedAt: time.Now().Unix(),
}
plaintext, errMarshal := json.Marshal(data)
if errMarshal != nil {
return "", fmt.Errorf("marshal compaction capsule: %w", errMarshal)
}
key := deriveAntigravityCompactionKey()
block, errCipher := aes.NewCipher(key)
if errCipher != nil {
return "", fmt.Errorf("create cipher: %w", errCipher)
}
gcm, errGCM := cipher.NewGCM(block)
if errGCM != nil {
return "", fmt.Errorf("create gcm: %w", errGCM)
}
nonce := make([]byte, gcm.NonceSize())
if _, errRead := io.ReadFull(rand.Reader, nonce); errRead != nil {
return "", fmt.Errorf("generate nonce: %w", errRead)
}
ciphertext := gcm.Seal(nonce, nonce, plaintext, nil)
return antigravityCompactionCapsulePrefix + base64.RawURLEncoding.EncodeToString(ciphertext), nil
}
// UnsealAntigravityCompaction decrypts and validates an opaque capsule.
func UnsealAntigravityCompaction(encryptedContent string) (string, error) {
if !strings.HasPrefix(encryptedContent, antigravityCompactionCapsulePrefix) {
return "", fmt.Errorf("unrecognized compaction capsule format")
}
raw := strings.TrimPrefix(encryptedContent, antigravityCompactionCapsulePrefix)
ciphertext, errDecode := base64.RawURLEncoding.DecodeString(raw)
if errDecode != nil {
return "", fmt.Errorf("decode compaction capsule: %w", errDecode)
}
key := deriveAntigravityCompactionKey()
block, errCipher := aes.NewCipher(key)
if errCipher != nil {
return "", fmt.Errorf("create cipher: %w", errCipher)
}
gcm, errGCM := cipher.NewGCM(block)
if errGCM != nil {
return "", fmt.Errorf("create gcm: %w", errGCM)
}
nonceSize := gcm.NonceSize()
if len(ciphertext) < nonceSize+gcm.Overhead() {
return "", fmt.Errorf("compaction capsule ciphertext too short")
}
nonce, encrypted := ciphertext[:nonceSize], ciphertext[nonceSize:]
plaintext, errOpen := gcm.Open(nil, nonce, encrypted, nil)
if errOpen != nil {
return "", fmt.Errorf("invalid or corrupted compaction capsule: %w", errOpen)
}
var data antigravityCompactionCapsuleData
if errUnmarshal := json.Unmarshal(plaintext, &data); errUnmarshal != nil {
return "", fmt.Errorf("unmarshal compaction capsule: %w", errUnmarshal)
}
return data.Summary, nil
}
// ExpandAntigravityCompactionCapsules expands compaction items in input into developer context.
func ExpandAntigravityCompactionCapsules(payload []byte) ([]byte, error) {
input := gjson.GetBytes(payload, "input")
if !input.IsArray() {
return payload, nil
}
var expandedItems []string
changed := false
for _, item := range input.Array() {
if item.Get("type").String() == "compaction" {
encryptedContent := item.Get("encrypted_content").String()
summary, errUnseal := UnsealAntigravityCompaction(encryptedContent)
if errUnseal != nil {
return nil, fmt.Errorf("invalid compaction capsule: %w", errUnseal)
}
devMsg := []byte(`{"type":"message","role":"developer","content":[{"type":"input_text"}]}`)
devMsg, _ = sjson.SetBytes(devMsg, "content.0.text", "Context summary from previous turns:\n"+summary)
expandedItems = append(expandedItems, string(devMsg))
changed = true
continue
}
expandedItems = append(expandedItems, item.Raw)
}
if !changed {
return payload, nil
}
newInput := "[" + strings.Join(expandedItems, ",") + "]"
return sjson.SetRawBytes(payload, "input", []byte(newInput))
}
// ExtractAntigravitySummaryText extracts summary text from Antigravity or Responses response.
func ExtractAntigravitySummaryText(respPayload []byte) (string, error) {
// 1. OpenAI Responses format: iterate through output array and skip reasoning
output := gjson.GetBytes(respPayload, "output")
if output.IsArray() {
var textParts []string
for _, item := range output.Array() {
if item.Get("type").String() == "message" {
content := item.Get("content")
if content.IsArray() {
for _, part := range content.Array() {
partType := part.Get("type").String()
if partType == "output_text" || partType == "text" {
if t := part.Get("text").String(); t != "" {
textParts = append(textParts, t)
}
}
}
} else if content.Type == gjson.String && content.String() != "" {
textParts = append(textParts, content.String())
}
}
}
if len(textParts) > 0 {
return strings.Join(textParts, "\n"), nil
}
}
// 2. Antigravity Gemini format: response.candidates.0.content.parts or candidates.0.content.parts
candidates := gjson.GetBytes(respPayload, "response.candidates")
if !candidates.Exists() {
candidates = gjson.GetBytes(respPayload, "candidates")
}
if candidates.IsArray() && len(candidates.Array()) > 0 {
var textParts []string
parts := candidates.Array()[0].Get("content.parts")
if parts.IsArray() {
for _, part := range parts.Array() {
if part.Get("thought").Bool() {
continue
}
if t := part.Get("text").String(); t != "" {
textParts = append(textParts, t)
}
}
}
if len(textParts) > 0 {
return strings.Join(textParts, "\n"), nil
}
}
// 3. Claude format: content array
content := gjson.GetBytes(respPayload, "content")
if content.IsArray() {
var textParts []string
for _, block := range content.Array() {
if block.Get("type").String() == "text" {
if t := block.Get("text").String(); t != "" {
textParts = append(textParts, t)
}
}
}
if len(textParts) > 0 {
return strings.Join(textParts, "\n"), nil
}
}
// 4. OpenAI Chat format: choices.0.message.content
if text := gjson.GetBytes(respPayload, "choices.0.message.content").String(); text != "" {
return text, nil
}
return "", fmt.Errorf("no summary text found in upstream response")
}
// BuildAntigravityCompactionStreamChunks creates SSE frames for compaction stream response.
func BuildAntigravityCompactionStreamChunks(modelName, capsule string, inputTokens, outputTokens, totalTokens int) [][]byte {
now := time.Now().Unix()
responseID := fmt.Sprintf("resp_ag_compact_%d", time.Now().UnixNano())
itemID := fmt.Sprintf("cmp_ag_compact_%d", time.Now().UnixNano())
itemInProgress := []byte(`{"type":"compaction","status":"in_progress"}`)
itemInProgress, _ = sjson.SetBytes(itemInProgress, "id", itemID)
itemInProgress, _ = sjson.SetBytes(itemInProgress, "encrypted_content", capsule)
itemCompleted := []byte(`{"type":"compaction","status":"completed"}`)
itemCompleted, _ = sjson.SetBytes(itemCompleted, "id", itemID)
itemCompleted, _ = sjson.SetBytes(itemCompleted, "encrypted_content", capsule)
usageNode := []byte(`{"input_tokens_details":{"cached_tokens":0},"output_tokens_details":{"reasoning_tokens":0}}`)
usageNode, _ = sjson.SetBytes(usageNode, "input_tokens", inputTokens)
usageNode, _ = sjson.SetBytes(usageNode, "output_tokens", outputTokens)
usageNode, _ = sjson.SetBytes(usageNode, "total_tokens", totalTokens)
createdResponse := []byte(`{"object":"response","status":"in_progress","background":false,"error":null,"output":[]}`)
createdResponse, _ = sjson.SetBytes(createdResponse, "id", responseID)
createdResponse, _ = sjson.SetBytes(createdResponse, "created_at", now)
createdResponse, _ = sjson.SetBytes(createdResponse, "model", modelName)
inProgressResponse := createdResponse
completedResponse := []byte(`{"object":"response","status":"completed","background":false,"error":null}`)
completedResponse, _ = sjson.SetBytes(completedResponse, "id", responseID)
completedResponse, _ = sjson.SetBytes(completedResponse, "created_at", now)
completedResponse, _ = sjson.SetBytes(completedResponse, "completed_at", now)
completedResponse, _ = sjson.SetBytes(completedResponse, "model", modelName)
completedResponse, _ = sjson.SetRawBytes(completedResponse, "output", []byte("["+string(itemCompleted)+"]"))
completedResponse, _ = sjson.SetRawBytes(completedResponse, "usage", usageNode)
createdPayload := []byte(`{"type":"response.created","sequence_number":0}`)
createdPayload, _ = sjson.SetRawBytes(createdPayload, "response", createdResponse)
inProgressPayload := []byte(`{"type":"response.in_progress","sequence_number":1}`)
inProgressPayload, _ = sjson.SetRawBytes(inProgressPayload, "response", inProgressResponse)
addedPayload := []byte(`{"type":"response.output_item.added","sequence_number":2,"output_index":0}`)
addedPayload, _ = sjson.SetRawBytes(addedPayload, "item", itemInProgress)
donePayload := []byte(`{"type":"response.output_item.done","sequence_number":3,"output_index":0}`)
donePayload, _ = sjson.SetRawBytes(donePayload, "item", itemCompleted)
completedPayload := []byte(`{"type":"response.completed","sequence_number":4}`)
completedPayload, _ = sjson.SetRawBytes(completedPayload, "response", completedResponse)
frames := [][]byte{
buildSSEFrame("response.created", createdPayload),
buildSSEFrame("response.in_progress", inProgressPayload),
buildSSEFrame("response.output_item.added", addedPayload),
buildSSEFrame("response.output_item.done", donePayload),
buildSSEFrame("response.completed", completedPayload),
}
return frames
}
// BuildAntigravityCompactionResponse creates JSON response for non-stream response.compaction.
func BuildAntigravityCompactionResponse(modelName, capsule string, inputTokens, outputTokens, totalTokens int) []byte {
now := time.Now().Unix()
responseID := fmt.Sprintf("resp_ag_compact_%d", time.Now().UnixNano())
itemID := fmt.Sprintf("cmp_ag_compact_%d", time.Now().UnixNano())
itemJSON := []byte(`{"type":"compaction","status":"completed"}`)
itemJSON, _ = sjson.SetBytes(itemJSON, "id", itemID)
itemJSON, _ = sjson.SetBytes(itemJSON, "encrypted_content", capsule)
usageJSON := []byte(`{"input_tokens_details":{"cached_tokens":0},"output_tokens_details":{"reasoning_tokens":0}}`)
usageJSON, _ = sjson.SetBytes(usageJSON, "input_tokens", inputTokens)
usageJSON, _ = sjson.SetBytes(usageJSON, "output_tokens", outputTokens)
usageJSON, _ = sjson.SetBytes(usageJSON, "total_tokens", totalTokens)
respJSON := []byte(`{"object":"response.compaction","status":"completed"}`)
respJSON, _ = sjson.SetBytes(respJSON, "id", responseID)
respJSON, _ = sjson.SetBytes(respJSON, "created_at", now)
respJSON, _ = sjson.SetBytes(respJSON, "model", modelName)
respJSON, _ = sjson.SetRawBytes(respJSON, "output", []byte("["+string(itemJSON)+"]"))
respJSON, _ = sjson.SetRawBytes(respJSON, "usage", usageJSON)
return respJSON
}
// ExtractResponsesCompactionUsage extracts usage counters from an OpenAI Responses summary payload.
func ExtractResponsesCompactionUsage(payload []byte) (inputTokens, outputTokens, totalTokens, cachedTokens, reasoningTokens int) {
if detail, ok := ParseCodexUsage(payload); ok && (detail.TotalTokens > 0 || detail.InputTokens > 0 || detail.OutputTokens > 0) {
return int(detail.InputTokens), int(detail.OutputTokens), int(detail.TotalTokens), int(detail.CachedTokens), int(detail.ReasoningTokens)
}
parsed := ParseOpenAIUsage(payload)
return int(parsed.InputTokens), int(parsed.OutputTokens), int(parsed.TotalTokens), int(parsed.CachedTokens), int(parsed.ReasoningTokens)
}
// PatchResponsesCompactionStreamUsage updates usage counters on a compaction SSE chunk.
func PatchResponsesCompactionStreamUsage(chunk []byte, inputTokens, outputTokens, totalTokens, cachedTokens, reasoningTokens int) []byte {
const dataPrefix = "data: "
idx := bytes.Index(chunk, []byte(dataPrefix))
if idx < 0 {
return chunk
}
data := bytes.TrimSpace(chunk[idx+len(dataPrefix):])
path := ""
switch {
case gjson.GetBytes(data, "response.usage").Exists():
path = "response.usage"
case gjson.GetBytes(data, "usage").Exists():
path = "usage"
default:
return chunk
}
data, _ = sjson.SetBytes(data, path+".input_tokens", inputTokens)
data, _ = sjson.SetBytes(data, path+".output_tokens", outputTokens)
data, _ = sjson.SetBytes(data, path+".total_tokens", totalTokens)
data, _ = sjson.SetBytes(data, path+".input_tokens_details.cached_tokens", cachedTokens)
data, _ = sjson.SetBytes(data, path+".output_tokens_details.reasoning_tokens", reasoningTokens)
var buf bytes.Buffer
buf.Write(chunk[:idx])
buf.WriteString(dataPrefix)
buf.Write(data)
buf.WriteString("\n\n")
return buf.Bytes()
}
// ValidateResponsesCompactionSummaryStatus ensures the upstream summary response finished with status "completed".
func ValidateResponsesCompactionSummaryStatus(summaryPayload []byte) error {
status := strings.ToLower(strings.TrimSpace(gjson.GetBytes(summaryPayload, "status").String()))
if status == "" {
status = strings.ToLower(strings.TrimSpace(gjson.GetBytes(summaryPayload, "response.status").String()))
}
if status != "completed" {
if status == "" {
status = "missing"
}
return fmt.Errorf("compaction summary response status is %q, want \"completed\"", status)
}
if eventType := strings.ToLower(strings.TrimSpace(gjson.GetBytes(summaryPayload, "type").String())); eventType == "response.incomplete" || eventType == "response.failed" {
return fmt.Errorf("compaction summary response terminal event is %q", eventType)
}
for _, path := range []string{"incomplete_details", "response.incomplete_details", "error", "response.error"} {
if node := gjson.GetBytes(summaryPayload, path); node.Exists() && node.Type != gjson.Null {
return fmt.Errorf("compaction summary response contains %s", path)
}
}
if output := gjson.GetBytes(summaryPayload, "output"); output.IsArray() {
for _, item := range output.Array() {
itemStatus := strings.ToLower(strings.TrimSpace(item.Get("status").String()))
if itemStatus != "" && itemStatus != "completed" {
return fmt.Errorf("compaction summary output item status is %q, want \"completed\"", itemStatus)
}
}
}
return nil
}
// BuildResponsesCompactionResponseFromSummary seals summary text from an OpenAI Responses payload into a response.compaction payload.
func BuildResponsesCompactionResponseFromSummary(modelName string, summaryPayload []byte) ([]byte, error) {
if errStatus := ValidateResponsesCompactionSummaryStatus(summaryPayload); errStatus != nil {
return nil, errStatus
}
summaryText, errExtract := ExtractAntigravitySummaryText(summaryPayload)
if errExtract != nil {
return nil, fmt.Errorf("extract summary: %w", errExtract)
}
capsule, errSeal := SealAntigravityCompaction(summaryText, modelName)
if errSeal != nil {
return nil, fmt.Errorf("seal compaction capsule: %w", errSeal)
}
inputTokens, outputTokens, totalTokens, cachedTokens, reasoningTokens := ExtractResponsesCompactionUsage(summaryPayload)
respPayload := BuildAntigravityCompactionResponse(modelName, capsule, inputTokens, outputTokens, totalTokens)
respPayload, _ = sjson.SetBytes(respPayload, "usage.input_tokens_details.cached_tokens", cachedTokens)
respPayload, _ = sjson.SetBytes(respPayload, "usage.output_tokens_details.reasoning_tokens", reasoningTokens)
return respPayload, nil
}
// BuildResponsesCompactionStreamChunksFromResponse converts a sealed response.compaction payload into SSE stream chunks.
func BuildResponsesCompactionStreamChunksFromResponse(modelName string, compactionPayload []byte) ([][]byte, error) {
item := gjson.GetBytes(compactionPayload, "output.0")
capsule := item.Get("encrypted_content").String()
if item.Get("type").String() != "compaction" || capsule == "" {
return nil, fmt.Errorf("extract summary: compaction item missing")
}
inputTokens := int(gjson.GetBytes(compactionPayload, "usage.input_tokens").Int())
outputTokens := int(gjson.GetBytes(compactionPayload, "usage.output_tokens").Int())
totalTokens := int(gjson.GetBytes(compactionPayload, "usage.total_tokens").Int())
cachedTokens := int(gjson.GetBytes(compactionPayload, "usage.input_tokens_details.cached_tokens").Int())
reasoningTokens := int(gjson.GetBytes(compactionPayload, "usage.output_tokens_details.reasoning_tokens").Int())
chunks := BuildAntigravityCompactionStreamChunks(modelName, capsule, inputTokens, outputTokens, totalTokens)
for i := range chunks {
chunks[i] = PatchResponsesCompactionStreamUsage(chunks[i], inputTokens, outputTokens, totalTokens, cachedTokens, reasoningTokens)
}
return chunks, nil
}
func buildSSEFrame(event string, data []byte) []byte {
var buf bytes.Buffer
buf.WriteString("event: ")
buf.WriteString(event)
buf.WriteString("\ndata: ")
buf.Write(data)
buf.WriteString("\n\n")
return buf.Bytes()
}