mirror of
https://github.com/deepseek-ai/deepseek-harness
synced 2026-08-15 21:04:50 +00:00
feat(session-title): add fallback and model providers
This commit is contained in:
@@ -0,0 +1,26 @@
|
||||
# @deepseek-ai/dsh-session-title-first-message-llm
|
||||
|
||||
Optional `ctx.sessionTitle` provider that summarizes the first eligible human message through `ctx.llm`. It registers the `first-message` cadence, runs automatically only when a fresh non-fork session first creates its fallback, and attributes the result to that message's exact seq. An automatic failure retains the fallback and is retried only through `ctx.sessionTitle.refresh()`.
|
||||
|
||||
The plugin uses the complete required [shared LLM configuration](../session-title-llm/README.md#configuration). Omit both `provider` and `model` to inherit the exact route from the current logged main request, or set both to route title generation independently.
|
||||
|
||||
## Model Experience
|
||||
|
||||
### First-message title request
|
||||
|
||||
#### What the model sees
|
||||
|
||||
The title model receives the shared title instruction and a JSON array containing only the first eligible human message. Later prompts and inherited fork history do not trigger another automatic call.
|
||||
|
||||
#### Token effect
|
||||
|
||||
At most one automatic auxiliary request is made for a fresh session, bounded by `maxInputBytes` and `maxOutputTokens`; explicit refreshes may make additional calls. The main agent request gains zero tokens.
|
||||
|
||||
#### KV Cache effect
|
||||
|
||||
No main-request invalidation. The auxiliary request uses the configured or logged route and has provider-specific cache behavior.
|
||||
|
||||
## Known Limitations and Deferred Work
|
||||
|
||||
- The first message alone may cease to represent a long-running session; use the all-messages provider when later prompts should retitle it.
|
||||
- A fork keeps its inherited title and never runs this provider automatically, even when its seeded first message came from the parent.
|
||||
@@ -0,0 +1,37 @@
|
||||
{
|
||||
"name": "@deepseek-ai/dsh-session-title-first-message-llm",
|
||||
"description": "First-message LLM provider plugin for DeepSeek Harness session titles",
|
||||
"version": "0.0.1",
|
||||
"private": true,
|
||||
"type": "module",
|
||||
"main": "lib/index.js",
|
||||
"types": "lib/types/index.d.ts",
|
||||
"exports": {
|
||||
".": {
|
||||
"types": "./lib/types/index.d.ts",
|
||||
"default": "./lib/index.js"
|
||||
},
|
||||
"./package.json": "./package.json"
|
||||
},
|
||||
"files": ["lib/index.js", "lib/types/**/*.d.ts", "lib/types/**/*.d.ts.map", "src"],
|
||||
"license": "BSD-3-Clause",
|
||||
"peerDependencies": {
|
||||
"@deepseek-ai/dsh-llm": "^0.0.1",
|
||||
"@deepseek-ai/dsh-session-title": "^0.0.1",
|
||||
"@deepseek-ai/dsh-session-title-llm": "^0.0.1",
|
||||
"cordis": "^4.0.0-rc.7"
|
||||
},
|
||||
"dependencies": {
|
||||
"schemastery": "^3.18.0"
|
||||
},
|
||||
"devDependencies": {
|
||||
"@cordisjs/plugin-include": "workspace:^",
|
||||
"@cordisjs/plugin-loader": "workspace:^",
|
||||
"@deepseek-ai/dsh-llm": "workspace:^",
|
||||
"@deepseek-ai/dsh-llm-deepseek": "workspace:^",
|
||||
"@deepseek-ai/dsh-session": "workspace:^",
|
||||
"@deepseek-ai/dsh-session-title": "workspace:^",
|
||||
"@deepseek-ai/dsh-session-title-llm": "workspace:^",
|
||||
"cordis": "^4.0.0-rc.7"
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,40 @@
|
||||
/** First-human-message model provider for `ctx.sessionTitle`. */
|
||||
|
||||
import type { Context } from 'cordis'
|
||||
import z from 'schemastery'
|
||||
import {
|
||||
registerSessionTitleLlmProvider,
|
||||
SessionTitleLlmConfigFields,
|
||||
} from '@deepseek-ai/dsh-session-title-llm'
|
||||
import type { SessionTitleLlmConfig } from '@deepseek-ai/dsh-session-title-llm'
|
||||
|
||||
export const name = 'session-title-first-message-llm'
|
||||
export const inject = ['sessionTitle', 'llm']
|
||||
|
||||
/** Required LLM policy; this plugin adds no defaults. */
|
||||
export type Config = SessionTitleLlmConfig
|
||||
/** Loader schema shared with the all-messages provider. */
|
||||
/* jscpd:ignore-start -- Loader requires each plugin to export its own statically walkable schema; the field validators remain shared. */
|
||||
export const Config: z<Config> = z.object({
|
||||
targetWords: SessionTitleLlmConfigFields.targetWords,
|
||||
targetCjkCharacters: SessionTitleLlmConfigFields.targetCjkCharacters,
|
||||
maxInputBytes: SessionTitleLlmConfigFields.maxInputBytes,
|
||||
maxOutputTokens: SessionTitleLlmConfigFields.maxOutputTokens,
|
||||
timeoutMs: SessionTitleLlmConfigFields.timeoutMs,
|
||||
provider: SessionTitleLlmConfigFields.provider,
|
||||
model: SessionTitleLlmConfigFields.model,
|
||||
})
|
||||
/* jscpd:ignore-end */
|
||||
|
||||
/**
|
||||
* Register the first-message model provider.
|
||||
* @param ctx - context exposing session-title and LLM services.
|
||||
* @param config - required route, target, byte, token, and timeout policy.
|
||||
*/
|
||||
export function apply(ctx: Context, config: Config): void {
|
||||
registerSessionTitleLlmProvider(ctx, config, name, 'first-message', (messages) => {
|
||||
const first = messages[0]
|
||||
if (first === undefined) throw new Error('first-message title provider requires one human message')
|
||||
return [first]
|
||||
})
|
||||
}
|
||||
@@ -0,0 +1,120 @@
|
||||
import { afterEach, describe, expect, it } from 'vitest'
|
||||
import { Context } from 'cordis'
|
||||
import Loader from '@cordisjs/plugin-loader'
|
||||
import Include from '@cordisjs/plugin-include'
|
||||
import { mkdtemp, rm, writeFile } from 'node:fs/promises'
|
||||
import { tmpdir } from 'node:os'
|
||||
import { join } from 'node:path'
|
||||
import { pathToFileURL } from 'node:url'
|
||||
import LlmService, { LlmAdapter } from '@deepseek-ai/dsh-llm'
|
||||
import type { GenerateOptions, StreamChunk } from '@deepseek-ai/dsh-llm'
|
||||
import SessionStore, { SessionId } from '@deepseek-ai/dsh-session'
|
||||
import SessionTitleService from '@deepseek-ai/dsh-session-title'
|
||||
import * as providerPlugin from '@deepseek-ai/dsh-session-title-first-message-llm'
|
||||
|
||||
let root: string | undefined
|
||||
let context: Context | undefined
|
||||
|
||||
class LoaderAdapter extends LlmAdapter {
|
||||
readonly requests: GenerateOptions[] = []
|
||||
|
||||
override async * stream(options: GenerateOptions): AsyncIterable<StreamChunk> {
|
||||
this.requests.push(options)
|
||||
yield { type: 'text-delta', index: 0, text: 'Loader composed title' }
|
||||
yield { type: 'finish', reason: { kind: 'stop' } }
|
||||
}
|
||||
}
|
||||
|
||||
afterEach(async () => {
|
||||
await context?.fiber.dispose()
|
||||
context = undefined
|
||||
if (root !== undefined) await rm(root, { recursive: true, force: true })
|
||||
root = undefined
|
||||
})
|
||||
|
||||
async function loadComposition(): Promise<Context> {
|
||||
root = await mkdtemp(join(tmpdir(), 'dsh-title-loader-'))
|
||||
const configPath = join(root, 'cordis.yml')
|
||||
await writeFile(configPath, [
|
||||
"- name: '@deepseek-ai/dsh-llm'",
|
||||
"- name: '@deepseek-ai/dsh-session'",
|
||||
"- name: '@deepseek-ai/dsh-session-title'",
|
||||
' config:',
|
||||
' fallbackMaxWords: 5',
|
||||
' fallbackMaxBytes: 40',
|
||||
' maxTitleBytes: 80',
|
||||
"- name: '@deepseek-ai/dsh-session-title-first-message-llm'",
|
||||
' config:',
|
||||
' targetWords: 5',
|
||||
' targetCjkCharacters: 10',
|
||||
' maxInputBytes: 1000',
|
||||
' maxOutputTokens: 32',
|
||||
' timeoutMs: 1000',
|
||||
" provider: 'title-route'",
|
||||
" model: 'title-model'",
|
||||
'',
|
||||
].join('\n'))
|
||||
|
||||
context = new Context()
|
||||
context.baseUrl = pathToFileURL(root).href + '/'
|
||||
await context.plugin(Loader)
|
||||
context.loader.builtins.include = Include
|
||||
const modules = new Map<string, unknown>([
|
||||
['@deepseek-ai/dsh-llm', LlmService],
|
||||
['@deepseek-ai/dsh-session', SessionStore],
|
||||
['@deepseek-ai/dsh-session-title', SessionTitleService],
|
||||
['@deepseek-ai/dsh-session-title-first-message-llm', providerPlugin],
|
||||
])
|
||||
context.loader.internal = {
|
||||
version: 'v2',
|
||||
async import(specifier: string) {
|
||||
if (!modules.has(specifier)) throw new Error(`unexpected Loader import: ${specifier}`)
|
||||
return modules.get(specifier)
|
||||
},
|
||||
} as unknown as NonNullable<typeof context.loader.internal>
|
||||
await context.loader.create({
|
||||
name: 'cordis:include',
|
||||
config: { path: pathToFileURL(configPath).href },
|
||||
})
|
||||
await context.loader.await()
|
||||
return context
|
||||
}
|
||||
|
||||
describe('session-title Loader composition', () => {
|
||||
it('loads the service and one model provider with required deployment policy', async () => {
|
||||
const ctx = await loadComposition()
|
||||
const unloaded = [...ctx.loader.entries()]
|
||||
.filter(entry => entry.fiber === undefined && !entry.disabled)
|
||||
.map(entry => entry.options.name)
|
||||
expect(unloaded).toEqual([])
|
||||
|
||||
const adapter = new LoaderAdapter()
|
||||
ctx.llm.registerAdapter(['title-route'], adapter)
|
||||
const session = ctx.sessions.create(SessionId('loader-title'))
|
||||
session.append('turn/start', {
|
||||
turn: 1,
|
||||
trigger: { kind: 'message', source: { kind: 'user' } },
|
||||
})
|
||||
const message = session.append('user/message', {
|
||||
content: [{ type: 'text', text: 'Compose a title through Loader' }],
|
||||
source: { kind: 'user' },
|
||||
}, { surfaceOp: 'append' })
|
||||
await new Promise(resolve => setTimeout(resolve, 0))
|
||||
session.append('request/header', {
|
||||
header: { config: { provider: 'main-route', model: 'main-model' } },
|
||||
reason: 'initial',
|
||||
})
|
||||
await new Promise(resolve => setTimeout(resolve, 0))
|
||||
|
||||
expect(adapter.requests[0]).toMatchObject({ provider: 'title-route', model: 'title-model' })
|
||||
expect(ctx.sessionTitle.get(session)).toMatchObject({
|
||||
title: 'Loader composed title',
|
||||
messageSeqs: [message.seq],
|
||||
source: {
|
||||
kind: 'provider',
|
||||
provider: 'session-title-first-message-llm',
|
||||
model: { provider: 'title-route', model: 'title-model' },
|
||||
},
|
||||
})
|
||||
})
|
||||
})
|
||||
@@ -0,0 +1,59 @@
|
||||
import { afterEach, describe, expect, it } from 'vitest'
|
||||
import { Context } from 'cordis'
|
||||
import LlmService from '@deepseek-ai/dsh-llm'
|
||||
import * as LlmDeepSeek from '@deepseek-ai/dsh-llm-deepseek'
|
||||
import SessionStore, { SessionId } from '@deepseek-ai/dsh-session'
|
||||
import SessionTitleService from '@deepseek-ai/dsh-session-title'
|
||||
import * as FirstMessageTitleProvider from '@deepseek-ai/dsh-session-title-first-message-llm'
|
||||
|
||||
const contexts: Context[] = []
|
||||
|
||||
afterEach(async () => {
|
||||
await Promise.all(contexts.splice(0).map(ctx => ctx.fiber.dispose()))
|
||||
})
|
||||
|
||||
describe.skipIf(!process.env.DEEPSEEK_API_KEY)('first-message title provider with real DeepSeek API', () => {
|
||||
it('replaces the fallback with a short model title', async () => {
|
||||
const ctx = new Context()
|
||||
contexts.push(ctx)
|
||||
await ctx.plugin(LlmService)
|
||||
await ctx.plugin(LlmDeepSeek, { thinking: 'disabled' })
|
||||
await ctx.plugin(SessionStore)
|
||||
await ctx.plugin(SessionTitleService, {
|
||||
fallbackMaxWords: 5,
|
||||
fallbackMaxBytes: 40,
|
||||
maxTitleBytes: 80,
|
||||
})
|
||||
await ctx.plugin(FirstMessageTitleProvider, {
|
||||
targetWords: 5,
|
||||
targetCjkCharacters: 10,
|
||||
maxInputBytes: 4_096,
|
||||
maxOutputTokens: 64,
|
||||
timeoutMs: 60_000,
|
||||
provider: 'deepseek',
|
||||
model: 'deepseek-v4-flash',
|
||||
})
|
||||
const session = ctx.sessions.create(SessionId('real-title-provider'))
|
||||
session.append('turn/start', {
|
||||
turn: 1,
|
||||
trigger: { kind: 'message', source: { kind: 'user' } },
|
||||
})
|
||||
const message = session.append('user/message', {
|
||||
content: [{ type: 'text', text: 'Explain why append-only logs make session titles durable.' }],
|
||||
source: { kind: 'user' },
|
||||
}, { surfaceOp: 'append' })
|
||||
|
||||
const title = await ctx.sessionTitle.refresh(session)
|
||||
|
||||
expect(title).toMatchObject({
|
||||
messageSeqs: [message.seq],
|
||||
source: {
|
||||
kind: 'provider',
|
||||
provider: 'session-title-first-message-llm',
|
||||
model: { provider: 'deepseek', model: 'deepseek-v4-flash' },
|
||||
},
|
||||
})
|
||||
expect(title?.title.length).toBeGreaterThan(0)
|
||||
expect(Buffer.byteLength(title?.title ?? '', 'utf8')).toBeLessThanOrEqual(80)
|
||||
})
|
||||
})
|
||||
@@ -0,0 +1,86 @@
|
||||
import { Context } from 'cordis'
|
||||
import { describe, expect, it, vi } from 'vitest'
|
||||
import LlmService, { LlmAdapter } from '@deepseek-ai/dsh-llm'
|
||||
import type { GenerateOptions, StreamChunk } from '@deepseek-ai/dsh-llm'
|
||||
import SessionStore, { Session, SessionId } from '@deepseek-ai/dsh-session'
|
||||
import SessionTitleService, { type SessionTitleProvider } from '@deepseek-ai/dsh-session-title'
|
||||
import * as providerPlugin from '@deepseek-ai/dsh-session-title-first-message-llm'
|
||||
|
||||
class RecordingAdapter extends LlmAdapter {
|
||||
readonly requests: GenerateOptions[] = []
|
||||
|
||||
override async * stream(options: GenerateOptions): AsyncIterable<StreamChunk> {
|
||||
this.requests.push(options)
|
||||
yield { type: 'text-delta', index: 0, text: 'First-message model title' }
|
||||
yield { type: 'finish', reason: { kind: 'stop' } }
|
||||
}
|
||||
}
|
||||
|
||||
const TITLE_CONFIG = { fallbackMaxWords: 5, fallbackMaxBytes: 40, maxTitleBytes: 80 } as const
|
||||
const LLM_CONFIG = {
|
||||
targetWords: 5,
|
||||
targetCjkCharacters: 10,
|
||||
maxInputBytes: 1_000,
|
||||
maxOutputTokens: 32,
|
||||
timeoutMs: 1_000,
|
||||
provider: 'title-route',
|
||||
model: 'title-model',
|
||||
} as const
|
||||
|
||||
async function settle(): Promise<void> {
|
||||
await new Promise(resolve => setTimeout(resolve, 0))
|
||||
}
|
||||
|
||||
describe('first-message LLM title provider', () => {
|
||||
it('rejects an impossible empty provider request at its own boundary', async () => {
|
||||
const ctx = new Context()
|
||||
await ctx.plugin(LlmService)
|
||||
await ctx.plugin(SessionStore)
|
||||
await ctx.plugin(SessionTitleService, TITLE_CONFIG)
|
||||
let registered: SessionTitleProvider | undefined
|
||||
vi.spyOn(ctx.sessionTitle, 'register').mockImplementation((provider) => {
|
||||
registered = provider
|
||||
return () => undefined
|
||||
})
|
||||
providerPlugin.apply(ctx, LLM_CONFIG)
|
||||
|
||||
await expect(registered!.generate({
|
||||
session: new Session(SessionId('empty-first-provider')),
|
||||
messages: [],
|
||||
signal: new AbortController().signal,
|
||||
})).rejects.toThrow(/requires one human message/)
|
||||
})
|
||||
|
||||
it('always selects only the first eligible human message, including explicit refresh', async () => {
|
||||
const ctx = new Context()
|
||||
await ctx.plugin(LlmService)
|
||||
await ctx.plugin(SessionStore)
|
||||
await ctx.plugin(SessionTitleService, TITLE_CONFIG)
|
||||
const adapter = new RecordingAdapter()
|
||||
ctx.llm.registerAdapter(['title-route'], adapter)
|
||||
await ctx.plugin(providerPlugin, LLM_CONFIG)
|
||||
const session = ctx.sessions.create(SessionId('first-plugin'))
|
||||
session.append('turn/start', { turn: 1, trigger: { kind: 'message', source: { kind: 'user' } } })
|
||||
const first = session.append('user/message', {
|
||||
content: [{ type: 'text', text: 'first input' }], source: { kind: 'user' },
|
||||
}, { surfaceOp: 'append' })
|
||||
await settle()
|
||||
session.append('request/header', {
|
||||
header: { config: { provider: 'main', model: 'main-model' } }, reason: 'initial',
|
||||
})
|
||||
await settle()
|
||||
session.append('user/message', {
|
||||
content: [{ type: 'text', text: 'second input must be ignored' }], source: { kind: 'user' },
|
||||
}, { surfaceOp: 'append' })
|
||||
|
||||
await ctx.sessionTitle.refresh(session)
|
||||
|
||||
expect(adapter.requests).toHaveLength(2)
|
||||
for (const options of adapter.requests) {
|
||||
const content = options.messages[0]?.content[0]
|
||||
expect(content?.type === 'text' && content.text).toContain('first input')
|
||||
expect(content?.type === 'text' && content.text).not.toContain('second input must be ignored')
|
||||
}
|
||||
expect(ctx.sessionTitle.get(session)).toMatchObject({ messageSeqs: [first.seq] })
|
||||
})
|
||||
})
|
||||
@@ -0,0 +1,13 @@
|
||||
{
|
||||
"extends": "../../../tsconfig.base.json",
|
||||
"compilerOptions": { "rootDir": "src", "outDir": "lib/types" },
|
||||
"include": ["src"],
|
||||
"references": [
|
||||
{ "path": "../../../vendor/cosmokit" },
|
||||
{ "path": "../../../vendor/cordis" },
|
||||
{ "path": "../../../vendor/schemastery" },
|
||||
{ "path": "../../llm/llm" },
|
||||
{ "path": "../session-title" },
|
||||
{ "path": "../session-title-llm" }
|
||||
]
|
||||
}
|
||||
Reference in New Issue
Block a user