valknar/llmx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add: codex-mini-latest (#951)

Browse Source 
💽

---------

Co-authored-by: Trevor Creech <tcreech@openai.com>
This commit is contained in:
Fouad Matin
2025-05-16 08:04:00 -07:00
committed by GitHub
parent ce2ecbe72f
commit 070499f534
9 changed files with 141 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

108
codex-cli/src/utils/agent/agent-loop.ts
View File

@@ -8,6 +8,7 @@ import type {
ResponseItem,
ResponseCreateParams,
FunctionTool,
Tool,
} from "openai/resources/responses/responses.mjs";
import type { Reasoning } from "openai/resources.mjs";
@@ -84,7 +85,7 @@ type AgentLoopParams = {
onLastResponseId: (lastResponseId: string) => void;
};
const shellTool: FunctionTool = {
const shellFunctionTool: FunctionTool = {
type: "function",
name: "shell",
description: "Runs a shell command, and returns its output.",
@@ -108,6 +109,11 @@ const shellTool: FunctionTool = {
},
};
const localShellTool: Tool = {
//@ts-expect-error - waiting on sdk
type: "local_shell",
};
export class AgentLoop {
private model: string;
private provider: string;
@@ -461,6 +467,73 @@ export class AgentLoop {
return [outputItem, ...additionalItems];
}
private async handleLocalShellCall(
// eslint-disable-next-line @typescript-eslint/no-explicit-any
item: any,
): Promise<Array<ResponseInputItem>> {
// If the agent has been canceled in the meantime we should not perform any
// additional work. Returning an empty array ensures that we neither execute
// the requested tool call nor enqueue any follow‑up input items. This keeps
// the cancellation semantics intuitive for users – once they interrupt a
// task no further actions related to that task should be taken.
if (this.canceled) {
return [];
}
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const outputItem: any = {
type: "local_shell_call_output",
// `call_id` is mandatory – ensure we never send `undefined` which would
// trigger the "No tool output found…" 400 from the API.
call_id: item.call_id,
output: "no function found",
};
// We intentionally *do not* remove this `callId` from the `pendingAborts`
// set right away. The output produced below is only queued up for the
// *next* request to the OpenAI API – it has not been delivered yet. If
// the user presses ESC‑ESC (i.e. invokes `cancel()`) in the small window
// between queuing the result and the actual network call, we need to be
// able to surface a synthetic `function_call_output` marked as
// "aborted". Keeping the ID in the set until the run concludes
// successfully lets the next `run()` differentiate between an aborted
// tool call (needs the synthetic output) and a completed one (cleared
// below in the `flush()` helper).
// used to tell model to stop if needed
const additionalItems: Array<ResponseInputItem> = [];
if (item.action.type !== "exec") {
throw new Error("Invalid action type");
}
const args = {
cmd: item.action.command,
workdir: item.action.working_directory,
timeoutInMillis: item.action.timeout_ms,
};
const {
outputText,
metadata,
additionalItems: additionalItemsFromExec,
} = await handleExecCommand(
args,
this.config,
this.approvalPolicy,
this.additionalWritableRoots,
this.getCommandConfirmation,
this.execAbortController?.signal,
);
outputItem.output = JSON.stringify({ output: outputText, metadata });
if (additionalItemsFromExec) {
additionalItems.push(...additionalItemsFromExec);
}
return [outputItem, ...additionalItems];
}
public async run(
input: Array<ResponseInputItem>,
previousResponseId: string = "",
@@ -545,6 +618,11 @@ export class AgentLoop {
// `disableResponseStorage === true`.
let transcriptPrefixLen = 0;
let tools: Array<Tool> = [shellFunctionTool];
if (this.model.startsWith("codex")) {
tools = [localShellTool];
}
const stripInternalFields = (
item: ResponseInputItem,
): ResponseInputItem => {
@@ -648,6 +726,8 @@ export class AgentLoop {
if (
(item as ResponseInputItem).type === "function_call" ||
(item as ResponseInputItem).type === "reasoning" ||
//@ts-expect-error - waiting on sdk
(item as ResponseInputItem).type === "local_shell_call" ||
((item as ResponseInputItem).type === "message" &&
// eslint-disable-next-line @typescript-eslint/no-explicit-any
(item as any).role === "user")
@@ -748,7 +828,7 @@ export class AgentLoop {
store: true,
previous_response_id: lastResponseId || undefined,
}),
tools: [shellTool],
tools: tools,
// Explicitly tell the model it is allowed to pick whatever
// tool it deems appropriate. Omitting this sometimes leads to
// the model ignoring the available tools and responding with
@@ -968,7 +1048,10 @@ export class AgentLoop {
if (maybeReasoning.type === "reasoning") {
maybeReasoning.duration_ms = Date.now() - thinkingStart;
}
if (item.type === "function_call") {
if (
item.type === "function_call" ||
item.type === "local_shell_call"
) {
// Track outstanding tool call so we can abort later if needed.
// The item comes from the streaming response, therefore it has
// either `id` (chat) or `call_id` (responses) – we normalise
@@ -1091,7 +1174,11 @@ export class AgentLoop {
let reasoning: Reasoning | undefined;
if (this.model.startsWith("o")) {
reasoning = { effort: "high" };
if (this.model === "o3" || this.model === "o4-mini") {
if (
this.model === "o3" ||
this.model === "o4-mini" ||
this.model === "codex-mini-latest"
) {
reasoning.summary = "auto";
}
}
@@ -1130,7 +1217,7 @@ export class AgentLoop {
store: true,
previous_response_id: lastResponseId || undefined,
}),
tools: [shellTool],
tools: tools,
tool_choice: "auto",
});
@@ -1492,6 +1579,17 @@ export class AgentLoop {
// eslint-disable-next-line no-await-in-loop
const result = await this.handleFunctionCall(item);
turnInput.push(...result);
//@ts-expect-error - waiting on sdk
} else if (item.type === "local_shell_call") {
//@ts-expect-error - waiting on sdk
if (alreadyProcessedResponses.has(item.id)) {
continue;
}
//@ts-expect-error - waiting on sdk
alreadyProcessedResponses.add(item.id);
// eslint-disable-next-line no-await-in-loop
const result = await this.handleLocalShellCall(item);
turnInput.push(...result);
}
emitItem(item as ResponseItem);
}

2
codex-cli/src/utils/config.ts
View File

@@ -43,7 +43,7 @@ if (!isVitest) {
loadDotenv({ path: USER_WIDE_CONFIG_PATH });
}
export const DEFAULT_AGENTIC_MODEL = "o4-mini";
export const DEFAULT_AGENTIC_MODEL = "codex-mini-latest";
export const DEFAULT_FULL_CONTEXT_MODEL = "gpt-4.1";
export const DEFAULT_APPROVAL_MODE = AutoApprovalMode.SUGGEST;
export const DEFAULT_INSTRUCTIONS = "";

4
codex-cli/src/utils/model-info.ts
View File

@@ -19,6 +19,10 @@ export const openAiModelInfo = {
label: "o3 (2025-04-16)",
maxContextLength: 200000,
},
"codex-mini-latest": {
label: "codex-mini-latest",
maxContextLength: 200000,
},
"o4-mini": {
label: "o4 Mini",
maxContextLength: 200000,