feat(aircoding): AirCoding V2 baseline — deterministic multi-agent architecture
Forked from OpenCode v1.17.4 with multi-agent system: - 5 agents: aircoding, scheduler, worker, architect, reviewer - Deterministic DAG scheduling engine (coordinator_tick) - Tool whitelists as hard enforcement - AirCoding validation plugin - System prompt injection for routing - V1 requirements: C4 docs, ADR, AGENTS.md, debug-log.md - Design documents in docs/
This commit is contained in:
airlongdian
232
packages/opencode/src/acp/usage.ts
Normal file
232
packages/opencode/src/acp/usage.ts
Normal file
@@ -0,0 +1,232 @@
|
||||
import type { AgentSideConnection, Usage } from "@agentclientprotocol/sdk"
|
||||
import type { AssistantMessage as OpenCodeAssistantMessage, Message } from "@opencode-ai/sdk/v2"
|
||||
import { InstanceRef } from "@/effect/instance-ref"
|
||||
import { InstanceStore } from "@/project/instance-store"
|
||||
import { ProviderV2 } from "@opencode-ai/core/provider"
|
||||
import { ModelV2 } from "@opencode-ai/core/model"
|
||||
import { Provider } from "@/provider/provider"
|
||||
import { Context, Effect, Layer, SynchronizedRef } from "effect"
|
||||
|
||||
export type AssistantTokenCost = Pick<OpenCodeAssistantMessage, "cost" | "tokens">
|
||||
|
||||
export type AssistantMessage = AssistantTokenCost &
|
||||
Pick<OpenCodeAssistantMessage, "role"> &
|
||||
Partial<Pick<OpenCodeAssistantMessage, "providerID" | "modelID">>
|
||||
|
||||
export type SessionMessage = {
|
||||
readonly info: { readonly role: Message["role"] } | AssistantMessage
|
||||
}
|
||||
|
||||
export type MessagesInput = {
|
||||
readonly sessionID: string
|
||||
readonly directory: string
|
||||
}
|
||||
|
||||
export type SDK = {
|
||||
readonly session: {
|
||||
readonly messages: (
|
||||
parameters: { readonly sessionID: string; readonly directory: string },
|
||||
options: { readonly throwOnError: true },
|
||||
) => Promise<{ readonly data?: readonly SessionMessage[] | null }>
|
||||
}
|
||||
}
|
||||
|
||||
export interface MessageLoaderInterface {
|
||||
readonly messages: (input: MessagesInput) => Effect.Effect<readonly SessionMessage[], unknown>
|
||||
}
|
||||
|
||||
export interface ContextLimitLoaderInterface {
|
||||
readonly providers: (directory: string) => Effect.Effect<Record<ProviderV2.ID, Provider.Info>, unknown>
|
||||
}
|
||||
|
||||
export type UsageConnection = Pick<AgentSideConnection, "sessionUpdate">
|
||||
|
||||
export interface Interface {
|
||||
readonly buildUsage: (message: AssistantTokenCost) => Usage
|
||||
readonly latestAssistantMessage: (messages: readonly SessionMessage[]) => AssistantMessage | undefined
|
||||
readonly totalSessionCost: (messages: readonly SessionMessage[]) => number
|
||||
readonly contextLimit: (input: {
|
||||
readonly directory: string
|
||||
readonly providerID: ProviderV2.ID
|
||||
readonly modelID: ModelV2.ID
|
||||
}) => Effect.Effect<number | undefined>
|
||||
readonly sendUpdate: (input: {
|
||||
readonly connection: UsageConnection
|
||||
readonly sessionID: string
|
||||
readonly directory: string
|
||||
}) => Effect.Effect<void>
|
||||
}
|
||||
|
||||
export class MessageLoader extends Context.Service<MessageLoader, MessageLoaderInterface>()(
|
||||
"@opencode/ACPUsageMessageLoader",
|
||||
) {}
|
||||
|
||||
export class ContextLimitLoader extends Context.Service<ContextLimitLoader, ContextLimitLoaderInterface>()(
|
||||
"@opencode/ACPUsageContextLimitLoader",
|
||||
) {}
|
||||
|
||||
export class Service extends Context.Service<Service, Interface>()("@opencode/ACPUsage") {}
|
||||
|
||||
export function messageLoaderFromSDK(sdk: SDK): MessageLoaderInterface {
|
||||
return MessageLoader.of({
|
||||
messages: (input) =>
|
||||
Effect.promise(() =>
|
||||
sdk.session
|
||||
.messages({ sessionID: input.sessionID, directory: input.directory }, { throwOnError: true })
|
||||
.then((response) => response.data ?? []),
|
||||
),
|
||||
})
|
||||
}
|
||||
|
||||
export const messageLoaderLayer = (sdk: SDK) => Layer.succeed(MessageLoader, messageLoaderFromSDK(sdk))
|
||||
|
||||
export function buildUsage(message: AssistantTokenCost): Usage {
|
||||
const cachedReadTokens = message.tokens.cache.read
|
||||
const cachedWriteTokens = message.tokens.cache.write
|
||||
const thoughtTokens = message.tokens.reasoning
|
||||
|
||||
return {
|
||||
inputTokens: message.tokens.input,
|
||||
outputTokens: message.tokens.output,
|
||||
totalTokens: message.tokens.input + message.tokens.output + thoughtTokens + cachedReadTokens + cachedWriteTokens,
|
||||
...(thoughtTokens > 0 ? { thoughtTokens } : {}),
|
||||
...(cachedReadTokens > 0 ? { cachedReadTokens } : {}),
|
||||
...(cachedWriteTokens > 0 ? { cachedWriteTokens } : {}),
|
||||
}
|
||||
}
|
||||
|
||||
export function latestAssistantMessage(messages: readonly SessionMessage[]): AssistantMessage | undefined {
|
||||
return messages
|
||||
.filter((message): message is { readonly info: AssistantMessage } => message.info.role === "assistant")
|
||||
.at(-1)?.info
|
||||
}
|
||||
|
||||
export function totalSessionCost(messages: readonly SessionMessage[]): number {
|
||||
return messages
|
||||
.filter((message): message is { readonly info: AssistantMessage } => message.info.role === "assistant")
|
||||
.reduce((sum, message) => sum + message.info.cost, 0)
|
||||
}
|
||||
|
||||
export function findContextLimit(
|
||||
providers: Record<ProviderV2.ID, Provider.Info>,
|
||||
providerID: ProviderV2.ID,
|
||||
modelID: ModelV2.ID,
|
||||
): number | undefined {
|
||||
return providers[providerID]?.models[modelID]?.limit.context
|
||||
}
|
||||
|
||||
export const contextLimitLoaderLayer = Layer.effect(
|
||||
ContextLimitLoader,
|
||||
Effect.gen(function* () {
|
||||
const store = yield* InstanceStore.Service
|
||||
const provider = yield* Provider.Service
|
||||
|
||||
return ContextLimitLoader.of({
|
||||
providers: Effect.fn("ACPUsageContextLimitLoader.providers")(function* (directory) {
|
||||
const ctx = yield* store.load({ directory })
|
||||
return yield* Effect.gen(function* () {
|
||||
return yield* provider.list()
|
||||
}).pipe(Effect.provideService(InstanceRef, ctx))
|
||||
}),
|
||||
})
|
||||
}),
|
||||
)
|
||||
|
||||
export const layer = Layer.effect(
|
||||
Service,
|
||||
Effect.gen(function* () {
|
||||
const messageLoader = yield* MessageLoader
|
||||
const contextLimitLoader = yield* ContextLimitLoader
|
||||
const limits = yield* SynchronizedRef.make(new Map<string, Effect.Effect<number | undefined>>())
|
||||
|
||||
const cachedLimit = Effect.fnUntraced(function* (input: {
|
||||
readonly directory: string
|
||||
readonly providerID: ProviderV2.ID
|
||||
readonly modelID: ModelV2.ID
|
||||
}) {
|
||||
return yield* SynchronizedRef.modifyEffect(
|
||||
limits,
|
||||
Effect.fnUntraced(function* (items) {
|
||||
const key = `${input.directory}\u0000${input.providerID}\u0000${input.modelID}`
|
||||
const current = items.get(key)
|
||||
if (current) return [current, items] as const
|
||||
const next = yield* Effect.cached(
|
||||
contextLimitLoader.providers(input.directory).pipe(
|
||||
Effect.map((providers) => findContextLimit(providers, input.providerID, input.modelID)),
|
||||
Effect.catch((error) =>
|
||||
Effect.logError("failed to get providers for usage context limit", { error: error }).pipe(
|
||||
Effect.as(undefined),
|
||||
),
|
||||
),
|
||||
),
|
||||
)
|
||||
return [next, new Map(items).set(key, next)] as const
|
||||
}),
|
||||
)
|
||||
})
|
||||
|
||||
const contextLimit = Effect.fn("ACPUsage.contextLimit")(function* (input: {
|
||||
readonly directory: string
|
||||
readonly providerID: ProviderV2.ID
|
||||
readonly modelID: ModelV2.ID
|
||||
}) {
|
||||
return yield* yield* cachedLimit(input)
|
||||
})
|
||||
|
||||
const sendUpdate = Effect.fn("ACPUsage.sendUpdate")(function* (input: {
|
||||
readonly connection: UsageConnection
|
||||
readonly sessionID: string
|
||||
readonly directory: string
|
||||
}) {
|
||||
const messages = yield* messageLoader
|
||||
.messages({ sessionID: input.sessionID, directory: input.directory })
|
||||
.pipe(
|
||||
Effect.catch((error) =>
|
||||
Effect.logError("failed to fetch messages for usage update", { error: error }).pipe(Effect.as(undefined)),
|
||||
),
|
||||
)
|
||||
if (!messages) return
|
||||
|
||||
const message = latestAssistantMessage(messages)
|
||||
if (!message) return
|
||||
if (!message.providerID || !message.modelID) return
|
||||
|
||||
const size = yield* contextLimit({
|
||||
directory: input.directory,
|
||||
providerID: ProviderV2.ID.make(message.providerID),
|
||||
modelID: ModelV2.ID.make(message.modelID),
|
||||
})
|
||||
if (!size) return
|
||||
|
||||
yield* Effect.promise(() =>
|
||||
input.connection
|
||||
.sessionUpdate({
|
||||
sessionId: input.sessionID,
|
||||
update: {
|
||||
sessionUpdate: "usage_update",
|
||||
used: message.tokens.input + message.tokens.cache.read,
|
||||
size,
|
||||
cost: { amount: totalSessionCost(messages), currency: "USD" },
|
||||
},
|
||||
})
|
||||
.catch(() => {}),
|
||||
)
|
||||
})
|
||||
|
||||
return Service.of({
|
||||
buildUsage,
|
||||
latestAssistantMessage,
|
||||
totalSessionCost,
|
||||
contextLimit,
|
||||
sendUpdate,
|
||||
})
|
||||
}),
|
||||
)
|
||||
|
||||
export const defaultLayer = layer.pipe(
|
||||
Layer.provide(contextLimitLoaderLayer),
|
||||
Layer.provide(Provider.defaultLayer),
|
||||
Layer.provide(InstanceStore.defaultLayer),
|
||||
)
|
||||
|
||||
export * as UsageService from "./usage"
|
||||
Reference in New Issue
Block a user