mirror of
https://github.com/openai/codex.git
synced 2026-05-01 09:56:37 +00:00
Add model-controlled truncation for code mode results (#14258)
Summary - document that `@openai/code_mode` exposes `set_max_output_tokens_per_exec_call` and that `code_mode` truncates the final Rust-side output when the budget is exceeded - enforce the configured budget in the Rust tool runner, reusing truncation helpers so text-only outputs follow the unified-exec wrapper and mixed outputs still fit within the limit - ensure the new behavior is covered by a code-mode integration test and string spec update Testing - Not run (not requested)
This commit is contained in:
@@ -1621,7 +1621,7 @@ source: /[\s\S]+/
|
||||
enabled_tool_names.join(", ")
|
||||
};
|
||||
let description = format!(
|
||||
"Runs JavaScript in a Node-backed `node:vm` context. This is a freeform tool: send raw JavaScript source text (no JSON/quotes/markdown fences). Direct tool calls remain available while `code_mode` is enabled. Inside JavaScript, import nested tools from `tools.js`, for example `import {{ exec_command }} from \"tools.js\"` or `import {{ tools }} from \"tools.js\"`. `tools[name]` and identifier wrappers like `await shell(args)` remain available for compatibility when the tool name is a valid JS identifier. Nested tool calls resolve to their code-mode result values. Function tools require JSON object arguments. Freeform tools require raw strings. Use synchronous `add_content(value)` with a content item, content-item array, or string. Structured nested-tool results should be converted to text first, for example with `JSON.stringify(...)`. Only content passed to `add_content(value)` is surfaced back to the model. Enabled nested tools: {enabled_list}."
|
||||
"Runs JavaScript in a Node-backed `node:vm` context. This is a freeform tool: send raw JavaScript source text (no JSON/quotes/markdown fences). Direct tool calls remain available while `code_mode` is enabled. Inside JavaScript, import nested tools from `tools.js`, for example `import {{ exec_command }} from \"tools.js\"` or `import {{ tools }} from \"tools.js\"`. `tools[name]` and identifier wrappers like `await shell(args)` remain available for compatibility when the tool name is a valid JS identifier. Nested tool calls resolve to their code-mode result values. Import `set_max_output_tokens_per_exec_call` from `@openai/code_mode` to set the token budget used to truncate the final Rust-side result of the current `code_mode` execution; the default is `10000`. This guards the overall `code_mode` output, not individual nested tool invocations. When truncation happens, the final text uses the unified-exec style `Original token count:` / `Output:` wrapper and the usual `…N tokens truncated…` marker. Function tools require JSON object arguments. Freeform tools require raw strings. Use synchronous `add_content(value)` with a content item, content-item array, or string. Structured nested-tool results should be converted to text first, for example with `JSON.stringify(...)`. Only content passed to `add_content(value)` is surfaced back to the model. Enabled nested tools: {enabled_list}."
|
||||
);
|
||||
|
||||
ToolSpec::Freeform(FreeformTool {
|
||||
|
||||
Reference in New Issue
Block a user