feat(acp-example): add record/replay llm/stream plugin for snapshot tests

Introduces examples/acp-agent/src/llm-replay.ts, a function/namespace plugin
that installs a single llm/stream waterfall listener. In record mode it tees
the real model's StreamChunks into a per-scenario llm.json (flushed atomically
after EACH stream, since the snapshot subprocess is SIGKILLed and start.ts has
no disposal path). In replay mode it short-circuits the waterfall and serves
recorded streams back positionally — the Nth stream() call gets the Nth entry —
so a snapshot test can drive the real agent with no API key.

Each fixture entry is a discriminated record {chunks|throw|hang} so it can
replay BOTH branches of the LLM failure contract (throw from stream() vs a
finish-error chunk) plus cancellation. A throw entry carries the prefix chunks
emitted before the throw, replayed before the error, so a mid-stream failure
(partial output then STREAM_CLOSED) reproduces what the loop saw live.

Fail-loud on a missing or exhausted fixture (never a silent skip). Unit tests
drive the real LlmService waterfall (record tee, ordered replay, the three
entry kinds, partial-then-throw, fail-loud, event-driven abort, HMR-safety).
Broadens the unit vitest include to examples/*/tests and registers the plugin
+ snapshot tests as knip entries. Per docs/rfc/implemented/2026-06-19.
This commit is contained in:
Tianyi Cui
2026-06-19 01:10:30 +08:00
parent bef9386591
commit 1a1ce734ba
5 changed files with 463 additions and 1 deletions

View File

@@ -30,6 +30,10 @@ Add to your Zed `settings.json` under `agent_servers`:
The editor sets each session's `cwd` to the project it opens; the agent's bash tools run there (see the per-session `cwd` note in `packages/acp`), so the server does not need to be launched in the workspace.
## Snapshot tests (record-once / replay-deterministic)
This example is the home of the harness's **snapshot tests** — they boot this server as a real subprocess, drive it with a deterministic input script, and diff its normalized stdout transcript against a committed golden file. The model is made deterministic by `src/llm-replay.ts`, a function/namespace plugin that installs an `llm/stream` waterfall listener: in `record` mode it tees the real model's `StreamChunk`s into a per-scenario `llm.json` (flushed atomically after each call); in `replay` mode it short-circuits the waterfall and serves those chunks back, so replay needs no API key. Each fixture entry is a discriminated record — `{ kind: 'chunks' | 'throw' | 'hang' }` — so both LLM failure branches (throw vs. finish-error) and cancellation replay faithfully. See [docs/rfc/implemented/2026-06-19-acp-snapshot-tests.md](../../docs/rfc/implemented/2026-06-19-acp-snapshot-tests.md) for the full design.
## MVP limitations
The bridge supports N concurrent sessions per connection, each in its own workspace `cwd` (RFC 011). Remaining limits: text-only prompts, `additionalDirectories` rejected (a session operates in its single `cwd`), and the tool-permission gate is deferred (`TODO(rfc010-permission-gate)` — tools run with the executor's full authority). See `packages/acp/README.md` for the full contract.

View File

@@ -0,0 +1,195 @@
/**
* Record/replay LLM plugin for snapshot tests.
*
* Installs a single `llm/stream` waterfall listener that, in `record` mode,
* tees the real model's streamed {@link StreamChunk}s into a fixture file, and
* in `replay` mode short-circuits the waterfall (never calls `next()`) to yield
* previously-recorded streams deterministically. This is the seam that lets a
* snapshot test boot the real agent against a fixed model transcript with no
* API key — see docs/rfc/implemented/2026-06-19-acp-snapshot-tests.md.
*
* It lives in the example (not packages/) because it is example/test
* infrastructure with one consumer, exactly like echo-agent's `mock-llm.ts`;
* the capability-seams rule says not to split into a published package
* preemptively.
*
* Plugin export shape: named `name`/`inject`/`Config`/`apply`, NO default
* export (the cordis Loader's `unwrapExports` does `exports.default ?? exports`,
* so a stray default would drop the namespace — see docs/postmortem/0001).
*/
import { existsSync, readFileSync, renameSync, writeFileSync } from 'node:fs'
import type { Context } from 'cordis'
import type { GenerateOptions, StreamChunk } from '@deepseek-ai/dsh-llm'
import { LlmError, assertNever } from '@deepseek-ai/dsh-llm'
/**
* One recorded model call. A discriminated union (not a bare `StreamChunk[]`)
* so it can faithfully replay BOTH branches of the documented LLM failure
* contract — an adapter may THROW from `stream()` or end with a `finish` error
* chunk — plus a `hang` marker for cancellation scenarios (mirrors the
* `MockAdapter` `hang` support in packages/agent-loop/tests).
*
* A `throw` entry carries any `chunks` the adapter emitted BEFORE it threw, so
* a mid-stream transport failure (partial output then `STREAM_CLOSED`) replays
* the partial chunks first and only then throws — exactly what the agent loop
* saw live (it may already have emitted partial assistant chunks).
*/
export type ReplayEntry =
| { kind: 'chunks'; chunks: StreamChunk[] }
| { kind: 'throw'; chunks: StreamChunk[]; message: string; code: string; status?: number }
| { kind: 'hang' }
/** Resolved plugin configuration. */
export interface ReplayConfig {
/** `record` tees the real model to `file`; `replay` serves `file` back. */
mode: 'record' | 'replay'
/** Path to the per-scenario `llm.json` fixture. */
file: string
}
/**
* Read and validate a fixture file. Throws a clear, fail-loud error when the
* file is missing (the scenario was never recorded) or malformed — never
* silently returns an empty script, so a coverage hole can't masquerade as a
* passing replay.
*/
export function loadFixture(file: string): ReplayEntry[] {
if (!existsSync(file)) {
throw new Error(`llm-replay: fixture not found: ${file} — run \`pnpm run test:snapshot:record\` first`)
}
const parsed: unknown = JSON.parse(readFileSync(file, 'utf8'))
if (!Array.isArray(parsed)) {
throw new Error(`llm-replay: fixture is not a JSON array: ${file}`)
}
// The fixture round-trips StreamChunk through JSON; branded CallId fields
// deserialize as plain strings, which are structurally StreamChunk. We trust
// the file shape (it is committed and produced by record mode) rather than
// deep-validating every chunk.
return parsed as ReplayEntry[]
}
/** Atomically write the recorded entries to `file` (temp write + rename). */
function flushFixture(file: string, entries: ReplayEntry[]): void {
const tmp = `${file}.tmp-${process.pid}`
writeFileSync(tmp, `${JSON.stringify(entries, null, 2)}\n`, { encoding: 'utf8' })
renameSync(tmp, file)
}
/** Yield a recorded stream back, honoring abort like a real adapter. */
async function* replayEntry(entry: ReplayEntry, signal: AbortSignal | undefined): AsyncIterable<StreamChunk> {
switch (entry.kind) {
case 'chunks':
for (const chunk of entry.chunks) {
if (signal?.aborted) throw new Error('aborted')
yield chunk
}
return
case 'throw':
// Replay the THROW branch of the LLM contract: emit whatever the adapter
// streamed before it threw (so the loop sees the same partial output it
// saw live), then throw the recorded error (e.g. a provider 401, or a
// mid-stream STREAM_CLOSED after partial chunks).
for (const chunk of entry.chunks) {
if (signal?.aborted) throw new Error('aborted')
yield chunk
}
throw new LlmError(entry.message, entry.code, entry.status)
case 'hang':
// Replay a stream that stalls until cancelled (mirrors MockAdapter): one
// chunk, then wait for abort and surface it as the consumer expects.
yield { type: 'block-start', index: 0, blockType: 'text' }
yield { type: 'text-delta', index: 0, text: 'partial' }
await new Promise<void>((_resolve, reject) => {
if (signal?.aborted) { reject(new Error('aborted')); return }
signal?.addEventListener('abort', () => { reject(new Error('aborted')) }, { once: true })
})
return
default:
// Closed local union: an unknown kind means malformed (hand-edited or
// drifted) fixture data — fail loud with a runtime diagnostic.
return assertNever(entry, 'llm-replay fixture entry')
}
}
/**
* Install the record/replay `llm/stream` listener on `ctx`. Returns the
* listener disposer (so a fiber dispose removes it — HMR safety). Exported
* separately from {@link apply} so unit tests can drive it without the Loader
* or env vars.
*
* Replay is POSITIONAL: the Nth `stream()` call serves the Nth fixture entry.
* This is deterministic only with at most one model stream in flight at a time;
* the snapshot harness runs one ACP session per scenario to guarantee that. The
* cursor is advanced synchronously at listener-invocation time (not lazily
* inside the generator) so call ORDER, not iteration order, fixes the mapping.
*/
export function installLlmReplay(ctx: Context, config: ReplayConfig): () => void {
if (config.mode === 'replay') {
const entries = loadFixture(config.file)
let cursor = 0
return ctx.on('llm/stream', (_options: GenerateOptions, _next) => {
const index = cursor++
const entry: ReplayEntry | undefined = entries[index]
return (async function* () {
if (entry === undefined) {
throw new Error(
`llm-replay: fixture exhausted — requested model call #${index + 1} but ${config.file} has only ${entries.length}; re-record the scenario`,
)
}
yield* replayEntry(entry, _options.signal)
})()
})
}
// Record mode: delegate to the real adapter via next(), tee each chunk, and
// flush atomically after EACH completed stream — the subprocess is SIGKILLed
// by the test teardown and start.ts has no disposal path, so a dispose-time
// flush would never run (see the RFC).
const recorded: ReplayEntry[] = []
return ctx.on('llm/stream', (_options: GenerateOptions, next) => {
const inner = next()
return (async function* () {
const chunks: StreamChunk[] = []
try {
for await (const chunk of inner) {
chunks.push(chunk)
yield chunk
}
} catch (error) {
const code = error instanceof LlmError ? error.code : 'UNKNOWN'
const status = error instanceof LlmError ? error.status : undefined
const message = error instanceof Error ? error.message : String(error)
// Record the chunks emitted before the throw alongside the error, so
// replay reproduces the same partial output + failure.
const entry: ReplayEntry = status === undefined
? { kind: 'throw', chunks, message, code }
: { kind: 'throw', chunks, message, code, status }
recorded.push(entry)
flushFixture(config.file, recorded)
throw error
}
recorded.push({ kind: 'chunks', chunks })
flushFixture(config.file, recorded)
})()
})
}
export const name = 'llm-replay'
export const inject = ['llm']
export interface Config {
/** Override the mode; defaults to `$DSH_SNAPSHOT` (`record`) else `replay`. */
mode?: 'record' | 'replay'
/** Override the fixture path; defaults to `$DSH_SNAPSHOT_FILE`. */
file?: string
}
export function apply(ctx: Context, config: Config = {}): void {
const mode = config.mode ?? (process.env.DSH_SNAPSHOT === 'record' ? 'record' : 'replay')
const file = config.file ?? process.env.DSH_SNAPSHOT_FILE
if (file === undefined || file.length === 0) {
throw new Error('llm-replay: a fixture path is required (Config.file or $DSH_SNAPSHOT_FILE)')
}
installLlmReplay(ctx, { mode, file })
}

View File

@@ -0,0 +1,262 @@
import { mkdtempSync, rmSync, writeFileSync } from 'node:fs'
import { tmpdir } from 'node:os'
import { join } from 'node:path'
import { afterEach, beforeEach, describe, expect, it } from 'vitest'
import { Context } from 'cordis'
import LlmService, { GenerateOptions, LlmAdapter, LlmError, StreamChunk } from '@deepseek-ai/dsh-llm'
import { type ReplayEntry, installLlmReplay, loadFixture } from '../src/llm-replay.ts'
/**
* Unit tests for the record/replay llm/stream plugin. These drive the listener
* through the REAL LlmService waterfall (not a hand-rolled stub) so they verify
* the actual seam the snapshot harness depends on.
*/
const TEXT_SCRIPT: StreamChunk[] = [
{ type: 'block-start', index: 0, blockType: 'text' },
{ type: 'text-delta', index: 0, text: 'hi' },
{ type: 'block-end', index: 0, block: { type: 'text', text: 'hi' } },
{ type: 'usage', usage: { inputTokens: 1, outputTokens: 1 } },
{ type: 'finish', reason: { kind: 'stop' } },
]
/** A scripted adapter whose every call yields one of a list of scripts. */
class MultiScriptAdapter extends LlmAdapter {
calls = 0
constructor(private scripts: (StreamChunk[] | (() => never))[]) {
super()
}
async * stream(_options: GenerateOptions): AsyncIterable<StreamChunk> {
const script = this.scripts[this.calls++]
if (script === undefined) throw new Error('MultiScriptAdapter: script exhausted')
if (typeof script === 'function') return script() // throws (returns never)
yield* script
}
}
let dir: string
let file: string
beforeEach(() => {
dir = mkdtempSync(join(tmpdir(), 'llm-replay-spec-'))
file = join(dir, 'llm.json')
})
afterEach(() => {
rmSync(dir, { recursive: true, force: true })
})
async function drain(iter: AsyncIterable<StreamChunk>): Promise<StreamChunk[]> {
const out: StreamChunk[] = []
for await (const chunk of iter) out.push(chunk)
return out
}
describe('llm-replay record mode', () => {
it('tees the real stream unchanged and flushes one chunks-entry per call', async () => {
const ctx = new Context()
await ctx.plugin(LlmService)
ctx.llm.registerAdapter(['m'], new MultiScriptAdapter([TEXT_SCRIPT]))
installLlmReplay(ctx, { mode: 'record', file })
const seen = await drain(ctx.llm.stream({ model: 'm', messages: [] }))
expect(seen).toEqual(TEXT_SCRIPT) // consumer sees the real chunks unchanged
const fixture = loadFixture(file)
expect(fixture).toEqual([{ kind: 'chunks', chunks: TEXT_SCRIPT }])
})
it('flushes after EACH stream (durable without dispose)', async () => {
const ctx = new Context()
await ctx.plugin(LlmService)
ctx.llm.registerAdapter(['m'], new MultiScriptAdapter([TEXT_SCRIPT, TEXT_SCRIPT]))
installLlmReplay(ctx, { mode: 'record', file })
await drain(ctx.llm.stream({ model: 'm', messages: [] }))
expect(loadFixture(file)).toHaveLength(1) // already on disk, no dispose needed
await drain(ctx.llm.stream({ model: 'm', messages: [] }))
expect(loadFixture(file)).toHaveLength(2)
})
it('records a throw-entry then re-throws when the adapter throws', async () => {
const ctx = new Context()
await ctx.plugin(LlmService)
ctx.llm.registerAdapter(['m'], new MultiScriptAdapter([() => { throw new Error('boom') }]))
installLlmReplay(ctx, { mode: 'record', file })
await expect(drain(ctx.llm.stream({ model: 'm', messages: [] }))).rejects.toThrow('boom')
expect(loadFixture(file)).toEqual([{ kind: 'throw', chunks: [], message: 'boom', code: 'UNKNOWN' }])
})
it('records the partial chunks emitted before a mid-stream throw', async () => {
const partial: StreamChunk[] = [
{ type: 'block-start', index: 0, blockType: 'text' },
{ type: 'text-delta', index: 0, text: 'par' },
]
function* chunkThenThrow(): Generator<StreamChunk> {
yield* partial
throw new LlmError('connection dropped', 'STREAM_CLOSED')
}
// An adapter that streams two chunks, then throws mid-stream.
class MidStreamThrowAdapter extends LlmAdapter {
async * stream(_options: GenerateOptions): AsyncIterable<StreamChunk> {
yield* chunkThenThrow()
}
}
const ctx = new Context()
await ctx.plugin(LlmService)
ctx.llm.registerAdapter(['m'], new MidStreamThrowAdapter())
installLlmReplay(ctx, { mode: 'record', file })
const seen: StreamChunk[] = []
await expect((async () => {
for await (const c of ctx.llm.stream({ model: 'm', messages: [] })) seen.push(c)
})()).rejects.toThrow('connection dropped')
expect(seen).toEqual(partial) // consumer saw the partial output before the throw
expect(loadFixture(file)).toEqual([
{ kind: 'throw', chunks: partial, message: 'connection dropped', code: 'STREAM_CLOSED' },
])
})
})
describe('llm-replay replay mode', () => {
function writeFixture(entries: ReplayEntry[]): void {
writeFileSync(file, JSON.stringify(entries), 'utf8')
}
it('serves recorded chunks back in order, short-circuiting the adapter', async () => {
writeFixture([{ kind: 'chunks', chunks: TEXT_SCRIPT }])
const ctx = new Context()
await ctx.plugin(LlmService)
// No adapter registered for 'm' — replay must not reach it.
installLlmReplay(ctx, { mode: 'replay', file })
const seen = await drain(ctx.llm.stream({ model: 'm', messages: [] }))
expect(seen).toEqual(TEXT_SCRIPT)
})
it('serves the Nth call the Nth entry (positional)', async () => {
const second: StreamChunk[] = [
{ type: 'block-start', index: 0, blockType: 'text' },
{ type: 'text-delta', index: 0, text: 'two' },
{ type: 'finish', reason: { kind: 'stop' } },
]
writeFixture([{ kind: 'chunks', chunks: TEXT_SCRIPT }, { kind: 'chunks', chunks: second }])
const ctx = new Context()
await ctx.plugin(LlmService)
installLlmReplay(ctx, { mode: 'replay', file })
expect(await drain(ctx.llm.stream({ model: 'm', messages: [] }))).toEqual(TEXT_SCRIPT)
expect(await drain(ctx.llm.stream({ model: 'm', messages: [] }))).toEqual(second)
})
it('replays a throw-entry as an LlmError with the recorded code/status', async () => {
writeFixture([{ kind: 'throw', chunks: [], message: 'unauthorized', code: 'AUTH', status: 401 }])
const ctx = new Context()
await ctx.plugin(LlmService)
installLlmReplay(ctx, { mode: 'replay', file })
await expect(drain(ctx.llm.stream({ model: 'm', messages: [] }))).rejects.toMatchObject({
message: 'unauthorized',
code: 'AUTH',
status: 401,
})
})
it('replays a throw-entry preceded by its partial chunks', async () => {
const partial: StreamChunk[] = [
{ type: 'block-start', index: 0, blockType: 'text' },
{ type: 'text-delta', index: 0, text: 'par' },
]
writeFixture([{ kind: 'throw', chunks: partial, message: 'dropped', code: 'STREAM_CLOSED' }])
const ctx = new Context()
await ctx.plugin(LlmService)
installLlmReplay(ctx, { mode: 'replay', file })
const seen: StreamChunk[] = []
await expect((async () => {
for await (const c of ctx.llm.stream({ model: 'm', messages: [] })) seen.push(c)
})()).rejects.toThrow('dropped')
expect(seen).toEqual(partial) // partial output replayed before the throw
})
it('replays a hang-entry that surfaces abort when the signal fires', async () => {
writeFixture([{ kind: 'hang' }])
const ctx = new Context()
await ctx.plugin(LlmService)
installLlmReplay(ctx, { mode: 'replay', file })
const controller = new AbortController()
const iterator = ctx.llm.stream({ model: 'm', messages: [], signal: controller.signal })[Symbol.asyncIterator]()
// Deterministically consume the two pre-hang chunks (no sleep), then abort
// and assert the next pull rejects — event-driven, per the no-sleeps rule.
expect((await iterator.next()).value).toMatchObject({ type: 'block-start' })
expect((await iterator.next()).value).toMatchObject({ type: 'text-delta' })
controller.abort()
await expect(iterator.next()).rejects.toThrow('aborted')
})
it('fails loud when the fixture is missing', () => {
const ctx = new Context()
expect(() => installLlmReplay(ctx, { mode: 'replay', file: join(dir, 'absent.json') }))
.toThrow(/fixture not found/)
})
it('fails loud when the fixture is exhausted', async () => {
writeFixture([{ kind: 'chunks', chunks: TEXT_SCRIPT }])
const ctx = new Context()
await ctx.plugin(LlmService)
installLlmReplay(ctx, { mode: 'replay', file })
await drain(ctx.llm.stream({ model: 'm', messages: [] }))
await expect(drain(ctx.llm.stream({ model: 'm', messages: [] }))).rejects.toThrow(/exhausted/)
})
it('aborts mid-replay when the signal is already set', async () => {
writeFixture([{ kind: 'chunks', chunks: TEXT_SCRIPT }])
const ctx = new Context()
await ctx.plugin(LlmService)
installLlmReplay(ctx, { mode: 'replay', file })
const controller = new AbortController()
controller.abort()
await expect(drain(ctx.llm.stream({ model: 'm', messages: [], signal: controller.signal })))
.rejects.toThrow('aborted')
})
})
describe('llm-replay HMR safety', () => {
it('removes the waterfall listener when the owning fiber is disposed', async () => {
writeFileSync(file, JSON.stringify([{ kind: 'chunks', chunks: TEXT_SCRIPT }]), 'utf8')
const ctx = new Context()
await ctx.plugin(LlmService)
ctx.llm.registerAdapter(['m'], new MultiScriptAdapter([TEXT_SCRIPT, TEXT_SCRIPT]))
const fiber = await ctx.plugin(Object.assign((inner: Context) => {
installLlmReplay(inner, { mode: 'replay', file })
}, { inject: ['llm'] }))
// While installed, replay short-circuits to the fixture ('hi').
expect(await drain(ctx.llm.stream({ model: 'm', messages: [] }))).toEqual(TEXT_SCRIPT)
await fiber.dispose()
// After dispose, the listener is gone and the call reaches the real adapter
// (also TEXT_SCRIPT here) — proving the waterfall no longer intercepts.
const afterDispose = await drain(ctx.llm.stream({ model: 'm', messages: [] }))
expect(afterDispose).toEqual(TEXT_SCRIPT)
})
})
describe('loadFixture', () => {
it('throws on a non-array JSON fixture', () => {
writeFileSync(file, '{"not":"an array"}', 'utf8')
expect(() => loadFixture(file)).toThrow(/not a JSON array/)
})
it('reads back what was written', () => {
const entries: ReplayEntry[] = [{ kind: 'chunks', chunks: TEXT_SCRIPT }]
writeFileSync(file, JSON.stringify(entries), 'utf8')
expect(loadFixture(file)).toEqual(entries)
})
})