Files
llmx/codex-rs/core/src/models.rs
Michael Bolin f48dd99f22 feat: add support for OpenAI tool type, local_shell (#961)
The new `codex-mini-latest` model expects a new tool with `{"type":
"local_shell"}`. Its contract is similar to the existing `function` tool
with `"name": "shell"`, so this takes the `local_shell` tool call into
`ExecParams` and sends it through the existing
`handle_container_exec_with_params()` code path.

This also adds the following logic when adding the default set of tools
to a request:

```rust
let default_tools = if self.model.starts_with("codex") {
    &DEFAULT_CODEX_MODEL_TOOLS
} else {
    &DEFAULT_TOOLS
};
```

That is, if the model name starts with `"codex"`, we add `{"type":
"local_shell"}` to the list of tools; otherwise, we add the
aforementioned `shell` tool.

To test this, I ran the TUI with `-m codex-mini-latest` and verified
that it used the `local_shell` tool. Though I also had some entries in
`[mcp_servers]` in my personal `config.toml`. The `codex-mini-latest`
model seemed eager to try the tools from the MCP servers first, so I
have personally commented them out for now, so keep an eye out if you're
testing `codex-mini-latest`!

Perhaps we should include more details with `{"type": "local_shell"}` or
update the following:


fd0b1b0208/codex-rs/core/prompt.md

For reference, the corresponding change in the TypeScript CLI is
https://github.com/openai/codex/pull/951.
2025-05-16 14:38:08 -07:00

264 lines
8.8 KiB
Rust
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
use std::collections::HashMap;
use base64::Engine;
use serde::Deserialize;
use serde::Serialize;
use serde::ser::Serializer;
use crate::protocol::InputItem;
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ResponseInputItem {
Message {
role: String,
content: Vec<ContentItem>,
},
FunctionCallOutput {
call_id: String,
output: FunctionCallOutputPayload,
},
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ContentItem {
InputText { text: String },
InputImage { image_url: String },
OutputText { text: String },
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ResponseItem {
Message {
role: String,
content: Vec<ContentItem>,
},
Reasoning {
id: String,
summary: Vec<ReasoningItemReasoningSummary>,
},
LocalShellCall {
/// Set when using the chat completions API.
id: Option<String>,
/// Set when using the Responses API.
call_id: Option<String>,
status: LocalShellStatus,
action: LocalShellAction,
},
FunctionCall {
name: String,
// The Responses API returns the function call arguments as a *string* that contains
// JSON, not as an alreadyparsed object. We keep it as a raw string here and let
// Session::handle_function_call parse it into a Value. This exactly matches the
// Chat Completions + Responses API behavior.
arguments: String,
call_id: String,
},
// NOTE: The input schema for `function_call_output` objects that clients send to the
// OpenAI /v1/responses endpoint is NOT the same shape as the objects the server returns on the
// SSE stream. When *sending* we must wrap the string output inside an object that includes a
// required `success` boolean. The upstream TypeScript CLI does this implicitly. To ensure we
// serialize exactly the expected shape we introduce a dedicated payload struct and flatten it
// here.
FunctionCallOutput {
call_id: String,
output: FunctionCallOutputPayload,
},
#[serde(other)]
Other,
}
impl From<ResponseInputItem> for ResponseItem {
fn from(item: ResponseInputItem) -> Self {
match item {
ResponseInputItem::Message { role, content } => Self::Message { role, content },
ResponseInputItem::FunctionCallOutput { call_id, output } => {
Self::FunctionCallOutput { call_id, output }
}
}
}
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum LocalShellStatus {
Completed,
InProgress,
Incomplete,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum LocalShellAction {
Exec(LocalShellExecAction),
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct LocalShellExecAction {
pub command: Vec<String>,
pub timeout_ms: Option<u64>,
pub working_directory: Option<String>,
pub env: Option<HashMap<String, String>>,
pub user: Option<String>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "type", rename_all = "snake_case")]
pub enum ReasoningItemReasoningSummary {
SummaryText { text: String },
}
impl From<Vec<InputItem>> for ResponseInputItem {
fn from(items: Vec<InputItem>) -> Self {
Self::Message {
role: "user".to_string(),
content: items
.into_iter()
.filter_map(|c| match c {
InputItem::Text { text } => Some(ContentItem::InputText { text }),
InputItem::Image { image_url } => Some(ContentItem::InputImage { image_url }),
InputItem::LocalImage { path } => match std::fs::read(&path) {
Ok(bytes) => {
let mime = mime_guess::from_path(&path)
.first()
.map(|m| m.essence_str().to_owned())
.unwrap_or_else(|| "application/octet-stream".to_string());
let encoded = base64::engine::general_purpose::STANDARD.encode(bytes);
Some(ContentItem::InputImage {
image_url: format!("data:{};base64,{}", mime, encoded),
})
}
Err(err) => {
tracing::warn!(
"Skipping image {} could not read file: {}",
path.display(),
err
);
None
}
},
})
.collect::<Vec<ContentItem>>(),
}
}
}
/// If the `name` of a `ResponseItem::FunctionCall` is either `container.exec`
/// or shell`, the `arguments` field should deserialize to this struct.
#[derive(Deserialize, Debug, Clone, PartialEq)]
pub struct ShellToolCallParams {
pub command: Vec<String>,
pub workdir: Option<String>,
/// This is the maximum time in seconds that the command is allowed to run.
#[serde(rename = "timeout")]
// The wire format uses `timeout`, which has ambiguous units, so we use
// `timeout_ms` as the field name so it is clear in code.
pub timeout_ms: Option<u64>,
}
#[derive(Deserialize, Debug, Clone)]
pub struct FunctionCallOutputPayload {
pub content: String,
#[expect(dead_code)]
pub success: Option<bool>,
}
// The Responses API expects two *different* shapes depending on success vs failure:
// • success → output is a plain string (no nested object)
// • failure → output is an object { content, success:false }
// The upstream TypeScript CLI implements this by specialcasing the serialize path.
// We replicate that behavior with a manual Serialize impl.
impl Serialize for FunctionCallOutputPayload {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: Serializer,
{
// The upstream TypeScript CLI always serializes `output` as a *plain string* regardless
// of whether the function call succeeded or failed. The boolean is purely informational
// for local bookkeeping and is NOT sent to the OpenAI endpoint. Sending the nested object
// form `{ content, success:false }` triggers the 400 we are still seeing. Mirror the JS CLI
// exactly: always emit a bare string.
serializer.serialize_str(&self.content)
}
}
// Implement Display so callers can treat the payload like a plain string when logging or doing
// trivial substring checks in tests (existing tests call `.contains()` on the output). Display
// returns the raw `content` field.
impl std::fmt::Display for FunctionCallOutputPayload {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.write_str(&self.content)
}
}
impl std::ops::Deref for FunctionCallOutputPayload {
type Target = str;
fn deref(&self) -> &Self::Target {
&self.content
}
}
#[cfg(test)]
mod tests {
#![allow(clippy::unwrap_used)]
use super::*;
#[test]
fn serializes_success_as_plain_string() {
let item = ResponseInputItem::FunctionCallOutput {
call_id: "call1".into(),
output: FunctionCallOutputPayload {
content: "ok".into(),
success: None,
},
};
let json = serde_json::to_string(&item).unwrap();
let v: serde_json::Value = serde_json::from_str(&json).unwrap();
// Success case -> output should be a plain string
assert_eq!(v.get("output").unwrap().as_str().unwrap(), "ok");
}
#[test]
fn serializes_failure_as_string() {
let item = ResponseInputItem::FunctionCallOutput {
call_id: "call1".into(),
output: FunctionCallOutputPayload {
content: "bad".into(),
success: Some(false),
},
};
let json = serde_json::to_string(&item).unwrap();
let v: serde_json::Value = serde_json::from_str(&json).unwrap();
assert_eq!(v.get("output").unwrap().as_str().unwrap(), "bad");
}
#[test]
fn deserialize_shell_tool_call_params() {
let json = r#"{
"command": ["ls", "-l"],
"workdir": "/tmp",
"timeout": 1000
}"#;
let params: ShellToolCallParams = serde_json::from_str(json).unwrap();
assert_eq!(
ShellToolCallParams {
command: vec!["ls".to_string(), "-l".to_string()],
workdir: Some("/tmp".to_string()),
timeout_ms: Some(1000),
},
params
);
}
}