Propagate DeepSeek SSE token usage to /v1/responses and remove stale POW env docs

[CJackHwang]

Motivation

• Fix an observed gap where DeepSeek SSE snapshots (e.g. accumulated_token_usage / token_usage) were not consistently reflected in OpenAI-compatible /v1/responses usage fields.
• Address a PR review note about stale documentation that advertised an unimplemented DS2API_POW_CONCURRENCY env var.

Description

• Propagate upstream prompt/output usage into responses: added promptTokens tracking to responsesStreamRuntime, capture PromptTokens/OutputTokens from parsed SSE lines and apply them to the final usage object (recomputing total_tokens).
• Non-stream responses now apply upstream token_usage.prompt_tokens as input_tokens and apply output_tokens overrides when present, and always recompute total_tokens from the final values.
• Added regression tests in Go (internal/adapter/openai/stream_status_test.go) and Node (tests/node/chat-stream.test.js) to cover response BATCH snapshots with accumulated_token_usage and token_usage cases, and updated a prompt-build test to match current tool-call instruction wording.
• Removed the outdated DS2API_POW_CONCURRENCY entries from README.MD and README.en.md.

Testing

• Ran go test ./internal/adapter/openai -count=1 and it passed (ok).
• Ran the Node parser tests with node --test tests/node/chat-stream.test.js and they passed (ok).
• Ran the full unit suite via ./tests/scripts/run-unit-all.sh and it completed successfully (ok).
• Ran lint via ./scripts/lint.sh, which bootstrapped golangci-lint and reported 0 issues.

---

Codex Task

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
ds2api	Ready	Preview, Comment	Apr 7, 2026 2:59am

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 668b9c26bd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".