Tianyi Cui
5a8234643a
GenerateOptions.prefill had no production setter and both adapters
rejected it with LlmError('UNSUPPORTED') — its entire observable
behavior was two throws, each pinned by one adapter test. DeepSeek's
chat-prefix completion is a Beta feature on a base URL neither adapter
targets. ToolSchema.strict was threaded through defineTool, the
registry's schemas() allowlist, the deepseek wire mapping, a per-tool
payload-patching pass in the pi-ai adapter, and a tool-catalog render
row, yet no shipped tool set it and the internal endpoint story for
strict mode was never built.
Remove both fields end-to-end: the vocabulary in dsh-llm, the adapter
guards and wire branches, the dsh-tools threading, the tool-catalog
Strict row, the pinning tests, the core.md pastes, the adapter README
rows, and the cookbook line that used prefill as the UNSUPPORTED
example (now stated generically). The pi-ai payload fixup keeps the
half with a job: pi-ai stamps strict:false on every serialized tool,
so the fixup scrubs it unconditionally for wire parity with the
hand-rolled twin (per-tool set/delete machinery gone). temperature/
stop/maxTokens are untouched — honored end-to-end by both adapters.
Each knob returns with its first real producer: prefill with an
adapter that implements chat-prefix completion, strict with a tool
that wants it and a beta-endpoint story.
RFC: docs/rfc/implemented/simplification/2026-07-04-drop-inert-request-knobs.md
(moved from proposed/, amended to shipped reality); the content-block
vocabulary RFC's consequence line now records prefill as producer-gated.
@deepseek-ai/dsh-llm-deepseek
DeepSeek chat-completions adapter for the harness LLM seam: hand-rolled fetch + SSE translation from the official wire format (source of truth: the API docs — guides/thinking_mode, guides/tool_calls, api/create-chat-completion) into the StreamChunk protocol.
A second, independent implementation of the same seam exists in @deepseek-ai/dsh-llm-pi-ai (library-backed). Same Config shape — pick one per context (registering both for the same model names throws by design).
Config
- id: llm-deepseek
name: '@deepseek-ai/dsh-llm-deepseek'
config:
apiKey: !!js process.env.DEEPSEEK_API_KEY # or rely on the env fallback
baseURL: !!js process.env.DEEPSEEK_BASE_URL # default: https://api.deepseek.com
models: [deepseek-v4-flash, deepseek-v4-pro] # one adapter, registered for each name
thinking: enabled # optional; provider default is enabled
reasoningEffort: high # optional; high | max — omitted ⇒ not sent
models lists every model name this one adapter instance serves: the adapter registers itself for each (the harness model name IS the wire model string), so a generate/stream call routes to it whenever options.model is any of them. Registering a second adapter for a name already taken throws LlmError('DUPLICATE_ADAPTER') (the LLM service enforces one adapter per model, all-or-nothing).
reasoningEffort is omitted by default — when unset, the reasoning_effort wire field is not sent and the server applies its own default for the model. The only accepted values are high and max (DeepSeek's official effort levels). It is meaningful only with thinking enabled (the provider default).
thinking/reasoningEffort are adapter-level request defaults serialized as the official top-level thinking: {type} / reasoning_effort wire fields. They live in adapter config (not GenerateOptions) to keep the core vocabulary provider-neutral.
Wire-format notes (verified live + against the official docs)
- Streaming only (
stream_options.include_usagealways on).usagemay arrive attached to the finish chunk or as a trailing usage-only chunk — the translator defers both to[DONE], sousagealways precedesfinishand nothing followsfinish. - The first thinking-mode chunk carries
reasoning_content: ""— handled (no spurious reasoning block). - Reasoning passback rule: on assistant turns that carried tool calls,
reasoning_contentis serialized back in history (required by the API in thinking mode); on tool-call-free turns it is dropped (ignored anyway — saves tokens). - Cache accounting:
cacheReadTokens←prompt_cache_hit_tokens/prompt_tokens_details.cached_tokens; DeepSeek reports no cache-write metric.
Limitations (MVP, documented deliberately)
tool_choiceis not mapped (not part of the core vocabulary).
Errors
Non-2xx responses throw LlmError with stable codes: AUTH (401/403), RATE_LIMIT (429), INVALID_REQUEST (400), SERVER (5xx), HTTP_<status> otherwise. Protocol violations throw STREAM_CLOSED (no [DONE]) or MALFORMED_RESPONSE (bad JSON payload). Unknown wire finish_reasons (e.g. content_filter, insufficient_system_resource) become finish {kind: 'error', code: <REASON>} chunks.
Testing
Unit suites run against a local node:http mock SSE server (no network). Real-API coverage lives in tests/adapter.e2e.ts (pnpm run test:e2e, key-gated): V4 Flash + V4 Pro across thinking enabled/disabled and both official effort levels, including the thinking+tools round trip with reasoning passback.