Merge latest origin/master into parallel-tool-call
This commit is contained in:
Tianyi Cui
@@ -11,7 +11,7 @@ Compaction extends [`SessionEventMap`](session.md) with three event types via de
|
||||
| Event | Payload | Role |
|
||||
|---|---|---|
|
||||
| `compact/start` | `{ turn }` | acquires the log-recorded lock |
|
||||
| `compact/summary` | `{ summary, shadowedRange, shadowedSeqs, shadowedTokenCount, model, maxTokens? }` | provenance: the summary blocks, the shadowed surface-boundary pair (`start`/`end` seqs — a position span, not a numeric interval), the shadowed seqs in surface order, the estimated token count, and the summarize call's envelope (`model`, plus its generation cap when one applied) — logged so the one-shot request is reconstructable from log + code (the reconstructability RFC) |
|
||||
| `compact/summary` | `{ summary, shadowedRange, shadowedSeqs, shadowedTokenCount, provider, model, maxTokens? }` | provenance: the summary blocks, the shadowed surface-boundary pair (`start`/`end` seqs — a position span, not a numeric interval), the shadowed seqs in surface order, the estimated token count, and the summarize call's envelope (`provider`, `model`, plus its generation cap when one applied) — logged so the one-shot request is reconstructable from log + code (the reconstructability RFC) |
|
||||
| `compact/end` | `{ turn, error? }` | releases the lock (`error` set when summarization threw) |
|
||||
|
||||
The lock brackets the **whole** operation: `compact/start` is appended first, then summarization, the `compact/summary` provenance record, and the `user/message` replacement all land, and only then `compact/end`. Releasing the lock last turns a crash mid-operation into a detectable orphaned lock (a `compact/start` with no matching `compact/end`) rather than a `compact/end` that falsely claims compaction finished.
|
||||
|
||||
@@ -104,12 +104,29 @@ interface ContentBlockMap {
|
||||
|
||||
The block interfaces (full fields in source): `TextBlock` (`text`), `ReasoningBlock` (thinking, distinct from visible text), `ToolCallBlock` (`id: CallId`, `name`, raw-JSON `arguments`), `ToolResultBlock` (`toolCallId`, nested `content: ContentBlock[]`, `isError?`). `ContentBlock = ContentBlockMap[ContentBlockType]`. The core set is limited to blocks every shipping path honors — multimodal content (images, audio, …) has no core block type; a feature that needs one adds it via the merge-extensible map together with the adapter/UI/compaction support that honors it.
|
||||
|
||||
A `Message` is a role plus blocks:
|
||||
A `Message` is a role plus blocks. Loop-derived assistant messages carry their durable provider/model identity and optional adapter-private replay metadata:
|
||||
|
||||
```ts type-equiv
|
||||
interface AssistantProvenance {
|
||||
/** Provider route that produced the message. */
|
||||
provider: string
|
||||
/** Provider model id that produced the message. */
|
||||
model: string
|
||||
/**
|
||||
* Lossless-JSON adapter state needed to replay the provider response.
|
||||
* `LlmService` exposes it to a target adapter only when that adapter instance
|
||||
* currently owns both this historical provider and the target provider.
|
||||
*/
|
||||
replayState?: unknown
|
||||
}
|
||||
```
|
||||
|
||||
```ts type-equiv
|
||||
interface Message {
|
||||
role: 'system' | 'user' | 'assistant'
|
||||
content: ContentBlock[]
|
||||
/** Present only on assistant messages produced by a routed adapter. */
|
||||
provenance?: AssistantProvenance
|
||||
}
|
||||
```
|
||||
|
||||
@@ -134,8 +151,34 @@ One model call is a fully-assembled `GenerateOptions`. The adapter answers with
|
||||
|
||||
Source: [`packages/llm/llm/src/types.ts`](../../packages/llm/llm/src/types.ts)
|
||||
|
||||
Provider and model discovery uses small provider-neutral descriptors. A model catalog is advisory: routing still keys on a registered provider, and an adapter may accept unlisted model ids.
|
||||
|
||||
```ts type-equiv
|
||||
interface LlmProviderInfo {
|
||||
/** Provider route key used by {@link GenerateOptions.provider}. */
|
||||
id: string
|
||||
/** Human-readable provider name for selectors and diagnostics. */
|
||||
name: string
|
||||
}
|
||||
```
|
||||
|
||||
```ts type-equiv
|
||||
interface LlmModelInfo {
|
||||
/** Provider route that owns this model entry. */
|
||||
provider: string
|
||||
/** Model id passed to {@link GenerateOptions.model}. */
|
||||
id: string
|
||||
/** Human-readable model name for selectors. */
|
||||
name: string
|
||||
/** Optional user-facing distinction from otherwise similar models. */
|
||||
description?: string
|
||||
}
|
||||
```
|
||||
|
||||
```ts type-equiv
|
||||
interface GenerateOptions {
|
||||
/** Registered provider route selecting the adapter instance. */
|
||||
provider: string
|
||||
model: string
|
||||
/**
|
||||
* Ordered conversation messages, exactly as the provider sees them (after
|
||||
@@ -196,7 +239,7 @@ The model-facing `ToolSchema` is the wire shape; the registered `ToolDefinition`
|
||||
|
||||
The loop builds each request from logged state. `EpochHeader` records call config, rendered prompt, authoritative returned tool order (configured by `toolOrder`, or lexicographic when unset), and session prefix through `request/header` snapshots and deltas. Together with derived history, this makes the request reconstructable from the session log. See [session.md](session.md#the-request-header-events-requestheader-and-requestheader-delta) and the [reconstructability RFC](../rfc/implemented/architecture/2026-07-05-reconstructable-requests.md).
|
||||
|
||||
`agent/request` receives a frozen call-config seed and may return a replacement. `agent/session-prefix` composes request-only prefix messages once per loop instance, and the header records the exact result used. Requests reaching `llm/stream` are deep-frozen, so mutation throws.
|
||||
`agent/request` receives a frozen call-config seed and may return a replacement to switch provider, model, or sampling. `agent/session-prefix` composes request-only prefix messages once per loop instance, and the header records the exact result used. Requests reaching `llm/stream` are deep-frozen, so mutation throws.
|
||||
|
||||
On the wire, a loop-built request reads in this order: the `system` slot (the rendered prompt assembly) → `messagePrefix` (the frozen session prefix) → the derived history — the boundary snapshot, whose tail is the newest `user/message` on a turn's first step and the previous step's tool results on later steps. The prefix never enters the derived history; its durable record is the header events, and the dev invariant recomputes exactly this equation against every loop-built request.
|
||||
|
||||
@@ -204,6 +247,7 @@ FIXME(call-config-shape): revisit the exact definition of this type — which fi
|
||||
|
||||
```ts type-equiv
|
||||
interface LlmCallConfig {
|
||||
provider: string
|
||||
model: string
|
||||
temperature?: number
|
||||
maxTokens?: number
|
||||
@@ -359,7 +403,7 @@ interface Agent {
|
||||
}
|
||||
```
|
||||
|
||||
`AgentStatus` is `'idle' | 'running' | 'disposed'`, and `AgentId` is branded. `AgentOptions` is merge-extensible and currently includes `model?`. Persona belongs to `dsh-system-prompt`: an agent-scoped `deployment:persona` may shadow the global default.
|
||||
`AgentStatus` is `'idle' | 'running' | 'disposed'`, and `AgentId` is branded. `AgentOptions` is merge-extensible and currently includes `provider?` and `model?`; dispatch requires both after `agent/request`. Persona belongs to `dsh-system-prompt`: an agent-scoped `deployment:persona` may shadow the global default.
|
||||
|
||||
The [event taxonomy](../architecture.md#event) owns the `agent/*` lifecycle, checkpoint, and waterfall contracts. Turn and step boundaries are durable session events rather than agent emits.
|
||||
|
||||
|
||||
@@ -16,7 +16,12 @@ type StreamChunk =
|
||||
| { type: 'tool-call-delta'; index: number; id: CallId; name?: string; argumentsDelta: string }
|
||||
| { type: 'block-end'; index: number; block: ContentBlock }
|
||||
| { type: 'usage'; usage: TokenUsage }
|
||||
| { type: 'finish'; reason: FinishReason }
|
||||
| {
|
||||
type: 'finish'
|
||||
reason: FinishReason
|
||||
/** Adapter-private lossless-JSON state for replaying a successful response. */
|
||||
replayState?: unknown
|
||||
}
|
||||
```
|
||||
|
||||
## The adapter contract
|
||||
@@ -27,8 +32,9 @@ Every adapter MUST obey these, and every consumer may rely on them:
|
||||
- **Tool-call `arguments` stay raw JSON strings end-to-end.** Partial fragments stream via `argumentsDelta`; a provider that hands back parsed objects re-stringifies at `block-end`.
|
||||
- **Two sanctioned error paths.** A failure may either THROW from `stream()` (transport/protocol errors) **or** end the stream with `finish {kind:'error'|'aborted'}` (provider in-band errors, for adapters that can't throw mid-stream). Consumers must handle *both*. The agent loop translates a finish-error/aborted into a turn error — it never logs a normal completed assistant message for a failed step.
|
||||
- **Every provider HTTP request carries the app-attribution header.** Adapters send `attributionHeaders()` (below) - the `User-Agent` baseline - and prove it with a wire-level test (mock server asserting the received header, or the library's header hook for a library-backed adapter).
|
||||
- **Replay state is adapter-owned.** A successful `finish` may carry lossless-JSON state needed to reconstruct a native provider response. The loop stores it with the assembled assistant message unless an `agent/step-result` listener rewrote the content. On a later request, `LlmService` passes the state only when the historical provider and target provider are currently registered to the exact same adapter instance. That adapter validates the state and owns any cross-model or cross-provider conversion; other adapters receive the provider-neutral content and provenance without the private state.
|
||||
|
||||
This contract is why two adapters exist as a deliberate pair: `dsh-llm-deepseek` (hand-rolled fetch/SSE) and `dsh-llm-pi-ai` (the same endpoint through `@earendil-works/pi-ai`). Two independent internals over one contract is what pinned the protocol down — the library-backed adapter can't throw mid-stream, so it exercises the finish-chunk error path the hand-rolled one might not.
|
||||
This contract was pinned down by two deliberately independent implementations: `dsh-llm-deepseek` (hand-rolled fetch/SSE) and `dsh-llm-pi-ai` (a generic multi-provider adapter through `@earendil-works/pi-ai`). The library-backed adapter cannot throw mid-stream, so it exercises the finish-chunk error path the hand-rolled one might not.
|
||||
|
||||
## `AppIdentity` — app attribution
|
||||
|
||||
@@ -58,11 +64,11 @@ interface TokenUsage {
|
||||
|
||||
## `BlockAssembler`
|
||||
|
||||
`BlockAssembler` ([`packages/llm/llm/src/assembler.ts`](../../packages/llm/llm/src/assembler.ts)) is the single shared implementation that folds a `StreamChunk` stream back into `ContentBlock`s and a final `Message`. The loop logs the raw chunks (for replay fidelity) while feeding the same chunks through an assembler — so the canonical log keeps token-level detail and the derived message is rebuilt deterministically. A consumer that needs the assembled result without re-implementing the fold uses this.
|
||||
`BlockAssembler` ([`packages/llm/llm/src/assembler.ts`](../../packages/llm/llm/src/assembler.ts)) is the single shared implementation that folds a `StreamChunk` stream back into `ContentBlock`s, usage, finish reason, and replay state. The loop logs the raw chunks while feeding the same chunks through an assembler, then stores the assembled assistant content with its provider/model provenance. A consumer that needs the assembled result without re-implementing the fold uses this.
|
||||
|
||||
## The seam
|
||||
|
||||
`LlmAdapter` is the provider seam: subclass, implement `stream()`, register with `ctx.llm.registerAdapter(models, adapter)`. The `block-start` / `block-end` `index` correlation and the assembler together mean an adapter only has to emit well-formed chunks — block reassembly is not each adapter's problem. The consumer surface (`ctx.llm.stream()`) and the `llm/stream` waterfall are described in [architecture.md § Content blocks and streaming](../architecture.md#content-blocks-and-streaming-dsh-llm).
|
||||
`LlmAdapter` is the provider seam: subclass, implement `stream()`, and register one adapter instance with `ctx.llm.registerAdapter(providers, adapter)`. `GenerateOptions.provider` selects the registered adapter; `GenerateOptions.model` is passed to that adapter and need not be registered at lifecycle start. Duplicate provider routes fail atomically. Optional `providerInfo()` and asynchronous `listModels()` methods feed `LlmService.listProviders()` / `listModels()` with detached selector metadata. That catalog is advisory rather than a request whitelist: the adapter remains authoritative and may accept unlisted model ids. Adapter lookup happens at the terminal continuation of the `llm/stream` waterfall, so a listener may short-circuit the call or route a mutable one-shot request before lookup. The `block-start` / `block-end` `index` correlation and the assembler together mean an adapter only has to emit well-formed chunks — block reassembly is not each adapter's problem. The consumer surface (`ctx.llm.stream()`) and the `llm/stream` waterfall are described in [architecture.md § Content blocks and streaming](../architecture.md#content-blocks-and-streaming-dsh-llm).
|
||||
|
||||
`ContentBlockType` (the key set the `index`-correlated blocks carry) derives from `ContentBlockMap`:
|
||||
|
||||
|
||||
@@ -56,7 +56,7 @@ interface SessionEventMap {
|
||||
* the model output and its accounting travel together (there is no separate
|
||||
* usage record). `usage` is absent when the adapter reported none.
|
||||
*/
|
||||
'assistant/message': { turn: number; step: number; content: ContentBlock[]; usage?: TokenUsage }
|
||||
'assistant/message': { turn: number; step: number; content: ContentBlock[]; provenance: AssistantProvenance; usage?: TokenUsage }
|
||||
'tool/call': { turn: number; step: number; callId: CallId; name: string; arguments: string }
|
||||
'tool/result': { turn: number; step: number; callId: CallId; content: ContentBlock[]; isError: boolean; error?: { name: string; code: string }; meta?: unknown }
|
||||
/** Steering content injected between steps of a running turn. */
|
||||
@@ -118,7 +118,7 @@ The request envelope — the `EpochHeader` (call config + rendered system prompt
|
||||
|
||||
```ts type-equiv
|
||||
export interface EpochHeader {
|
||||
/** The conversation's call configuration (model + sampling scalars). */
|
||||
/** The conversation's call configuration (provider + model + sampling scalars). */
|
||||
config: LlmCallConfig
|
||||
/** Rendered system prompt text; absent for a system-less request. */
|
||||
system?: string
|
||||
@@ -238,12 +238,12 @@ export interface SurfaceFoldResult {
|
||||
`Session.deriveMessages()` projects the event log into the `Message[]` the model sees — cached (each surface node projected once, when first seen; a surface rewrite rebuilds) and frozen (a fresh array per call over shared, deep-frozen messages, so mutating logged history through a projection is unrepresentable). `deriveEventMessage(event)` is the per-node pure function the fold applies — public so external reconstructors and the dev invariant project a log prefix with exactly the same rules and cannot disagree with the cache. The projection rules:
|
||||
|
||||
- `user/message` → a user message.
|
||||
- `assistant/message` → an assistant message. Raw `assistant/chunk` events are replay/UI data and are **skipped** in derivation (the assembled message is authoritative). An **empty-content** `assistant/message` is also skipped — a max-tokens step cut off with no content still records an `assistant/message` to host its `usage`, but a content-less assistant turn must not enter the provider transcript.
|
||||
- `assistant/message` → an assistant message with the event's provider/model provenance and optional adapter-private replay state. Raw `assistant/chunk` events are replay/UI data and are **skipped** in derivation (the assembled message is authoritative). An **empty-content** `assistant/message` is also skipped — a max-tokens step cut off with no content still records an `assistant/message` to host its usage/provenance, but a content-less assistant turn must not enter the provider transcript.
|
||||
- `tool/result` → a user message carrying a `tool-result` block.
|
||||
- `context/message` → a user-role message at its chronological position. The default `envelope` is `context`, which wraps content as `<context source="…">…</context>`; `envelope: 'raw'` uses caller-owned framing verbatim. Optional JSON `meta` remains in the event log and is never rendered.
|
||||
- `steering/message` → a user-role message wrapped in `<steering source="…">…</steering>` at its chronological position.
|
||||
|
||||
Everything else (`turn/*`, `step/*`) is structural and does not project into a message. Token usage is observed on `assistant/message.usage` (the step that produced it); an operational error's step number is on `turn/end.reason` for `kind: 'error'`.
|
||||
Everything else (`turn/*`, `step/*`) is structural and does not project into a message. Token usage is observed on `assistant/message.usage` (the step that produced it); an operational error's step number is on `turn/end.reason` for `kind: 'error'`. Because this unreleased format intentionally has no compatibility promise, seed/load validation rejects request headers without provider+model and assistant messages without provider/model provenance instead of guessing a route for historical data.
|
||||
|
||||
## Live-session fork API
|
||||
|
||||
|
||||
Reference in New Issue
Block a user