mirror of https://github.com/deepseek-ai/deepseek-harness synced 2026-08-15 21:04:50 +00:00

Files

_Kerman cb2f01f48c refactor(agent-loop): drop the steering/message session event

Steer, inject, and followup now land as durable user/message events on the
session surface; the steering/message event type and its ConversationNode
kind are removed from the client projection. Update tests, docs, generated
catalogs, and agent notes to match, and align the steering e2e fixture and
prompt inventory assertions with the durable user/message landing.

2026-08-03 16:04:26 +08:00

2.6 KiB

Raw Blame History

Agent Note: Provider-neutral content-block vocabulary owned by dsh-llm

Status: implemented

English | 中文

Problem

The harness needs one internal language for messages that the loop, session log, and all plugins speak.

Decision

Own the vocabulary: messages are arrays of typed content blocks (text, reasoning, tool-call, tool-result), with the union derived from the merge-extensible ContentBlockMap so plugins add block types via declaration merging. The same merge-extensible-map pattern types every "stringly" field (MessageSource, FinishReason, TurnTrigger, TurnEndReason). Streaming is a raw chunk protocol; BlockAssembler is the single shared assembly implementation. Adapters translate to provider wire formats — mapping cost lives in adapters, where it belongs.

In-session context injection (context/message) and mid-turn steering originally rendered as tagged user-role envelopes (the system-reminder pattern) rather than a new role, so adapters carry zero burden. Both now project as plain user content with no wrapper; see the injected-content-envelope Agent Note. Live-adapter validation confirms this rendering for current DeepSeek behavior; a future provider-specific mismatch belongs in that adapter rather than a new canonical role.

Alternatives considered

Mirror the DeepSeek/OpenAI chat-completions shape — zero mapping cost for the first provider, but awkward for rich content (reasoning, tool results as structured blocks).
Adopt Anthropic's Messages block structure verbatim — battle-tested, but the canonical types would mirror a third-party API the harness does not target first.

Consequences

Reasoning has a core home without provider-specific shapes.
Multimodal blocks return only with coordinated adapter, UI, and compaction support; see the drop-image Agent Note.
Cache hints and assistant prefill remain absent until a shipping adapter can honor them; see the producer-less variants and inert request knobs Agent Notes.
Every adapter pays a translation cost; the first real adapters have since validated the streaming protocol, and new adapters should continue proving their provider-specific mapping in adapter-local tests.
IDs that cross package boundaries are branded (CallId, the shared agent/session SessionId) — nominal typing at zero runtime cost.

2.6 KiB Raw Blame History