Skip to content

Add codemode: typechecked TypeScript execution with tools #16

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jonastemplestein wants to merge 14 commits into
base: main
Choose a base branch
from
Open

Add codemode: typechecked TypeScript execution with tools #16

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
2d1b26d
Add codemode infrastructure for code block execution
jonastemplestein Dec 4, 2025
3d24cc4
Add codemode CLI integration and E2E tests
jonastemplestein Dec 4, 2025
9923c43
Persist CodemodeResult events and include in LLM history
jonastemplestein Dec 4, 2025
1da5846
Add getSecret tool, agent loop, and codemode system prompt
jonastemplestein Dec 4, 2025
1a98e59
Always enable codemode (remove --codemode/-x flag)
jonastemplestein Dec 4, 2025
82f6d9a
Fix CodemodeResult import in generated code
jonastemplestein Dec 4, 2025
8b3a1e7
Handle typecheck failures in agent loop
jonastemplestein Dec 4, 2025
326abc2
Clarify codemode prompt: no imports needed, use t.log
jonastemplestein Dec 4, 2025
6dedc76
Replace endTurn with triggerAgentTurn and refactor codemode tools
jonastemplestein Dec 5, 2025
4339d91
Fix Scope requirement after rebase
jonastemplestein Dec 5, 2025
6f7a46a
Add standalone codemode run CLI command and refactor agent loop
jonastemplestein Dec 5, 2025
76713c1
Remove unused response.md logging from codemode repository
jonastemplestein Dec 5, 2025
2277f26
Remove test artifact files (example.txt, output.txt)
jonastemplestein Dec 5, 2025
8f32ec5
Fix codemode not working in tty-interactive mode
jonastemplestein Dec 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion eslint.config.mjs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@ const compat = new FlatCompat({

export default [
{
ignores: ["**/dist", "**/build", "**/node_modules", "**/*.md", ".contexts"]
ignores: ["**/dist", "**/build", "**/node_modules", "**/*.md", ".contexts", ".mini-agent/**"]
},
...compat.extends(
"eslint:recommended",
Expand Down
143 changes: 120 additions & 23 deletions src/cli/chat-ui.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,19 +6,18 @@
*/
import type { AiError, LanguageModel } from "@effect/ai"
import type { Error as PlatformError, FileSystem } from "@effect/platform"
import { Cause, Context, Effect, Fiber, Layer, Mailbox, Stream } from "effect"
import { is } from "effect/Schema"
import { Cause, Context, Effect, Fiber, Layer, Mailbox, Schema, Stream } from "effect"
import {
AssistantMessageEvent,
type ContextEvent,
CodemodeResultEvent,
CodemodeValidationErrorEvent,
LLMRequestInterruptedEvent,
TextDeltaEvent,
UserMessageEvent
} from "../context.model.ts"
import { ContextService } from "../context.service.ts"
import type { ContextLoadError, ContextSaveError } from "../errors.ts"
import { type ContextOrCodemodeEvent, ContextService } from "../context.service.ts"
import type { CodeStorageError, ContextLoadError, ContextSaveError } from "../errors.ts"
import type { CurrentLlmConfig } from "../llm-config.ts"
import { streamLLMResponse } from "../llm.ts"
import { type ChatController, runOpenTUIChat } from "./components/opentui-chat.tsx"

type ChatSignal =
Expand All @@ -32,7 +31,7 @@ export class ChatUI extends Context.Tag("@app/ChatUI")<
contextName: string
) => Effect.Effect<
void,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError | CodeStorageError,
LanguageModel.LanguageModel | FileSystem.FileSystem | CurrentLlmConfig
>
}
Expand Down Expand Up @@ -81,7 +80,7 @@ const runChatLoop = (
mailbox: Mailbox.Mailbox<ChatSignal>
): Effect.Effect<
void,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError | CodeStorageError,
LanguageModel.LanguageModel | FileSystem.FileSystem | CurrentLlmConfig
> =>
Effect.fn("ChatUI.runChatLoop")(function*() {
Expand All @@ -97,6 +96,18 @@ type TurnResult =
| { readonly _tag: "continue" }
| { readonly _tag: "exit" }

/** Check if event is displayable in the chat feed */
const isDisplayableEvent = (event: ContextOrCodemodeEvent): boolean =>
Schema.is(TextDeltaEvent)(event) ||
Schema.is(AssistantMessageEvent)(event) ||
Schema.is(CodemodeResultEvent)(event) ||
Schema.is(CodemodeValidationErrorEvent)(event)

/** Check if event triggers continuation (agent loop) */
const triggersContinuation = (event: ContextOrCodemodeEvent): boolean =>
(Schema.is(CodemodeResultEvent)(event) && event.triggerAgentTurn === "after-current-turn") ||
(Schema.is(CodemodeValidationErrorEvent)(event) && event.triggerAgentTurn === "after-current-turn")

const runChatTurn = (
contextName: string,
contextService: Context.Tag.Service<typeof ContextService>,
Expand All @@ -105,7 +116,7 @@ const runChatTurn = (
pendingMessage: string | null
): Effect.Effect<
TurnResult,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError | CodeStorageError,
LanguageModel.LanguageModel | FileSystem.FileSystem | CurrentLlmConfig
> =>
Effect.fn("ChatUI.runChatTurn")(function*() {
Expand All @@ -129,28 +140,105 @@ const runChatTurn = (
}

const userEvent = new UserMessageEvent({ content: userMessage })

yield* contextService.persistEvent(contextName, userEvent)
chat.addEvent(userEvent)

const events = yield* contextService.load(contextName)
let accumulatedText = ""
let needsContinuation = false

// Use contextService.addEvents with codemode enabled
const eventStream = contextService.addEvents(contextName, [userEvent], { codemode: true })

const streamFiber = yield* Effect.fork(
streamLLMResponse(events).pipe(
Stream.tap((event: ContextEvent) =>
eventStream.pipe(
Stream.tap((event: ContextOrCodemodeEvent) =>
Effect.sync(() => {
if (is(TextDeltaEvent)(event)) {
if (Schema.is(TextDeltaEvent)(event)) {
accumulatedText += event.delta
}
if (triggersContinuation(event)) {
needsContinuation = true
}
if (isDisplayableEvent(event)) {
chat.addEvent(event)
}
})
),
Stream.filter(is(AssistantMessageEvent)),
Stream.tap((event) =>
Effect.gen(function*() {
yield* contextService.persistEvent(contextName, event)
chat.addEvent(event)
Stream.runDrain
)
)

const result = yield* awaitStreamCompletion(streamFiber, mailbox)

if (result._tag === "completed") {
// If we need continuation (codemode result with output), run another turn
if (needsContinuation) {
return yield* runAgentContinuation(contextName, contextService, chat, mailbox)
}
return { _tag: "continue" } as const
}

if (result._tag === "exit") {
if (accumulatedText.length > 0) {
const interruptedEvent = new LLMRequestInterruptedEvent({
requestId: crypto.randomUUID(),
reason: "user_cancel",
partialResponse: accumulatedText
})
yield* contextService.persistEvent(contextName, interruptedEvent)
chat.addEvent(interruptedEvent)
}
return { _tag: "exit" } as const
}

// result._tag === "interrupted" - user hit return during streaming
if (accumulatedText.length > 0) {
const interruptedEvent = new LLMRequestInterruptedEvent({
requestId: crypto.randomUUID(),
reason: result.newMessage ? "user_new_message" : "user_cancel",
partialResponse: accumulatedText
})
yield* contextService.persistEvent(contextName, interruptedEvent)
chat.addEvent(interruptedEvent)
}

if (result.newMessage) {
return yield* runChatTurn(contextName, contextService, chat, mailbox, result.newMessage)
}

return { _tag: "continue" } as const
})()

/** Run agent continuation loop (for codemode results that need follow-up) */
const runAgentContinuation = (
contextName: string,
contextService: Context.Tag.Service<typeof ContextService>,
chat: ChatController,
mailbox: Mailbox.Mailbox<ChatSignal>
): Effect.Effect<
TurnResult,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError | CodeStorageError,
LanguageModel.LanguageModel | FileSystem.FileSystem | CurrentLlmConfig
> =>
Effect.fn("ChatUI.runAgentContinuation")(function*() {
let accumulatedText = ""
let needsContinuation = false

// Empty input events - the persisted CodemodeResult triggers the turn
const eventStream = contextService.addEvents(contextName, [], { codemode: true })

const streamFiber = yield* Effect.fork(
eventStream.pipe(
Stream.tap((event: ContextOrCodemodeEvent) =>
Effect.sync(() => {
if (Schema.is(TextDeltaEvent)(event)) {
accumulatedText += event.delta
}
if (triggersContinuation(event)) {
needsContinuation = true
}
if (isDisplayableEvent(event)) {
chat.addEvent(event)
}
})
),
Stream.runDrain
Expand All @@ -160,6 +248,9 @@ const runChatTurn = (
const result = yield* awaitStreamCompletion(streamFiber, mailbox)

if (result._tag === "completed") {
if (needsContinuation) {
return yield* runAgentContinuation(contextName, contextService, chat, mailbox)
Copy link

@cursor cursor bot Dec 6, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Bug: Missing iteration limit allows unbounded agent loop recursion

The runAgentContinuation function recursively calls itself at line 252 without any iteration limit, unlike the CLI's runEventStream which uses MAX_AGENT_LOOP_ITERATIONS = 15. This unbounded recursion occurs whenever needsContinuation is true, which happens when CodemodeResultEvent or CodemodeValidationErrorEvent has triggerAgentTurn === "after-current-turn". If the LLM consistently produces typechecking errors, outputs to stdout via console.log(), or fails to include <codemode> tags, the agent loop will recurse indefinitely, potentially causing stack overflow or resource exhaustion in the interactive chat UI.

Additional Locations (1)

Fix in Cursor Fix in Web

}
return { _tag: "continue" } as const
}

Expand All @@ -176,7 +267,7 @@ const runChatTurn = (
return { _tag: "exit" } as const
}

// result._tag === "interrupted" - user hit return during streaming
// Interrupted - save partial and return to wait for input
if (accumulatedText.length > 0) {
const interruptedEvent = new LLMRequestInterruptedEvent({
requestId: crypto.randomUUID(),
Expand All @@ -200,9 +291,15 @@ type StreamResult =
| { readonly _tag: "interrupted"; readonly newMessage: string | null }

const awaitStreamCompletion = (
fiber: Fiber.RuntimeFiber<void, AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError>,
fiber: Fiber.RuntimeFiber<
void,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError | CodeStorageError
>,
mailbox: Mailbox.Mailbox<ChatSignal>
): Effect.Effect<StreamResult, AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError> =>
): Effect.Effect<
StreamResult,
AiError.AiError | PlatformError.PlatformError | ContextLoadError | ContextSaveError | CodeStorageError
> =>
Effect.fn("ChatUI.awaitStreamCompletion")(function*() {
const waitForFiber = Fiber.join(fiber).pipe(Effect.as({ _tag: "completed" } as StreamResult))
const waitForInterrupt = Effect.gen(function*() {
Expand Down
Loading
Loading