- Narrow websocket transcript replacement detection to assistant outputs and function calls
- Preserve existing merge behavior for follow-up developer messages without previous_response_id
- Add a regression test covering mid-session developer message updates
- Add shouldReplaceWebsocketTranscript() to detect historical model output in input
- Add normalizeResponseTranscriptReplacement() for full transcript reset handling
- Prevent duplicate stale turn-state when clients replace local history post-compaction
- Avoid orphaned function_call items from incremental append on compact transcripts
- Add unit tests for transcript replacement detection and state reset behavior
Reverts the streaming thinking suppression introduced in b15453c.
rewriteStreamEvent should only inject signatures and rewrite model
names — suppressing thinking blocks in streaming mode breaks SSE
index alignment and causes the Amp TUI to render empty responses
on the second message onward (especially with model-mapped
non-Claude providers like GPT-5.4).
Non-streaming responses still suppress thinking when tool_use is
present via rewriteModelInResponse.
When a Claude assistant message contains [text, tool_use, text], the
Antigravity API internally splits the model message at functionCall
boundaries, creating an extra assistant turn between tool_use and the
following tool_result. Claude then rejects with:
tool_use ids were found without tool_result blocks immediately after
Fix: extend the existing 2-way part reordering (thinking-first) to a
3-way partition: thinking → regular → functionCall. This ensures
functionCall parts are always last, so Antigravity's split cannot
insert an extra assistant turn before the user's tool_result.
Fixes#989
- Call suppressAmpThinking in rewriteStreamEvent for streaming path
- Handle nil return from suppressAmpThinking to skip suppressed events
- Narrow looksLikeSSEChunk to line-prefix detection (HasPrefix vs Contains)
- Initialize suppressedContentBlock map in test
Drop the last affinity-related executor artifacts so the PR stays focused on the minimal Codex continuity fix set: stable prompt cache identity, stable session_id, and the executor-only behavior that was validated to restore cache reads.
Drop the chat-completions translator edits from this PR so the branch complies with the repository policy that forbids pull-request changes under internal/translator. The remaining PR stays focused on the executor-level Codex continuity fix that was validated to restore cache reuse.
Remove the experimental auth-affinity routing changes from this PR so it stays focused on the validated Codex continuity fix. This keeps the prompt-cache repair while avoiding unrelated routing-policy concerns such as provider/model affinity scope, lifecycle cleanup, and hard-pin fallback semantics.
Keep sticky auth affinity limited to matching providers and stop persisting execution-session IDs as long-lived affinity keys so provider switching and normal streaming traffic do not create incorrect pins or stale affinity state.
Stop using one-shot idempotency keys as long-lived auth-affinity identifiers and remove raw affinity-key values from debug logs so sticky routing keeps its continuity benefits without creating avoidable memory growth or credential exposure risks.
Restore Claude continuity after the continuity refactor, keep auth-affinity keys out of upstream Codex session identifiers, and only persist affinity after successful execution so retries can still rotate to healthy credentials when the first auth fails.
Align websocket continuity resolution with the HTTP Codex path, make auth-affinity principal keys use a stable string representation, and extract small helpers that remove duplicated continuity and affinity logic without changing the validated cache-hit behavior.
Prompt caching on Codex was not reliably reusable through the proxy because repeated chat-completions requests could reach the upstream without the same continuity envelope. In practice this showed up most clearly with OpenCode, where cache reads worked in the reference client but not through CLIProxyAPI, although the root cause is broader than OpenCode itself.
The proxy was breaking continuity in several ways: executor-layer Codex request preparation stripped prompt_cache_retention, chat-completions translation did not preserve that field, continuity headers used a different shape than the working client behavior, and OpenAI-style Codex requests could be sent without a stable prompt_cache_key. When that happened, session_id fell back to a fresh random value per request, so upstream Codex treated repeated requests as unrelated turns instead of as part of the same cacheable context.
This change fixes that by preserving caller-provided prompt_cache_retention on Codex execution paths, preserving prompt_cache_retention when translating OpenAI chat-completions requests to Codex, aligning Codex continuity headers to session_id, and introducing an explicit Codex continuity policy that derives a stable continuity key from the best available signal. The resolution order prefers an explicit prompt_cache_key, then execution session metadata, then an explicit idempotency key, then stable request-affinity metadata, then a stable client-principal hash, and finally a stable auth-ID hash when no better continuity signal exists.
The same continuity key is applied to both prompt_cache_key in the request body and session_id in the request headers so repeated requests reuse the same upstream cache/session identity. The auth manager also keeps auth selection sticky for repeated request sequences, preventing otherwise-equivalent Codex requests from drifting across different upstream auth contexts and accidentally breaking cache reuse.
To keep the implementation maintainable, the continuity resolution and diagnostics are centralized in a dedicated Codex continuity helper instead of being scattered across executor flow code. Regression coverage now verifies retention preservation, continuity-key precedence, stable auth-ID fallback, websocket parity, translator preservation, and auth-affinity behavior. Manual validation confirmed prompt cache reads now occur through CLIProxyAPI when using Codex via OpenCode, and the fix should also benefit other clients that rely on stable repeated Codex request continuity.
