Enhanced ChatGPT Clone: Features Agents, MCP, Skills, DeepSeek, Anthropic, AWS, OpenAI, Responses API, Azure, Groq, o1, GPT-5, Mistral, OpenRouter, Vertex AI, Gemini, Artifacts, AI model switching, message search, Code Interpreter, langchain, DALL-E-3, OpenAPI Actions, Functions, Secure Multi-User Auth, Presets, open-source for self-hosting. Active
TypeScript aianthropicartifactsawsazurechatgptchatgpt-cloneclaudeclonedeepseekgeminigooglegpt-5librechatmcpo1openairesponses-apivisionwebui
import { cleanCodeToolOutput } from './cleanup';
/**
* Real bash_tool outputs captured from a deployed conversation. Patterns
* tested live: trailing single `Note:` paragraph, double trailing
* `Note:` paragraphs, per-file `| <annotation>` suffixes (downloaded /
* displayed / known-as-input), and inline-comma file lists.
*/
describe('cleanCodeToolOutput', () => {
it('returns empty/falsy input unchanged', () => {
expect(cleanCodeToolOutput('')).toBe('');
});
it('strips a trailing single `Note:` paragraph (mkdir-only output)', () => {
const input = [
'stdout:',
'Folder created successfully!',
'',
'Generated files:',
'- /mnt/data/test_folder/.dirkeep',
'',
'',
'Note: Files from previous executions are automatically available and can be modified.',
].join('\n');
const output = cleanCodeToolOutput(input);
expect(output).not.toMatch(/Note:/);
expect(output).toMatch(/Folder created successfully!/);
expect(output).toMatch(/Generated files:/);
expect(output).toMatch(/- \/mnt\/data\/test_folder\/\.dirkeep$/);
});
it('strips both trailing `Note:` paragraphs when two are emitted', () => {
const input = [
'stdout:',
'File created successfully!',
'',
'Generated files:',
'- /mnt/data/test_folder/test_file.txt',
'Available files (inputs, not generated by this execution):',
'- /mnt/data/test_folder/.dirkeep',
'',
'',
'Note: Files from previous executions are automatically available and can be modified.',
'',
'Note: Files in "Available files" are inputs the user (or a skill) already provided to the sandbox. They were not produced by this execution and you should not present them as new outputs in your response.',
].join('\n');
const output = cleanCodeToolOutput(input);
expect(output).not.toMatch(/Note:/);
expect(output).toMatch(/Generated files:/);
expect(output).toMatch(/Available files \(inputs/);
});
it('strips per-file "downloaded" / "known to the user" / "displayed" annotations', () => {
const input = [
'Generated files:',
'- /mnt/data/a.txt | File is already downloaded by the user',
'- /mnt/data/b.png | Image is already displayed to the user',
'Available files (inputs, not generated by this execution):',
'- /mnt/data/c.txt | Available as an input — already known to the user',
].join('\n');
const output = cleanCodeToolOutput(input);
expect(output).not.toMatch(/already downloaded/);
expect(output).not.toMatch(/already displayed/);
expect(output).not.toMatch(/already known to the user/);
expect(output).toMatch(/- \/mnt\/data\/a\.txt$/m);
expect(output).toMatch(/- \/mnt\/data\/b\.png$/m);
expect(output).toMatch(/- \/mnt\/data\/c\.txt$/m);
});
it('handles inline comma-separated file lists (small file count formatting)', () => {
// The bash executor inlines the list with `, ` when ≤ 3 files.
const input =
'Available files (inputs, not generated by this execution):\n- /mnt/data/test_folder/test_file.txt | Available as an input — already known to the user, - /mnt/data/test_folder/.dirkeep | Available as an input — already known to the user';
const output = cleanCodeToolOutput(input);
expect(output).not.toMatch(/already known to the user/);
expect(output).toMatch(/- \/mnt\/data\/test_folder\/test_file\.txt/);
expect(output).toMatch(/- \/mnt\/data\/test_folder\/\.dirkeep/);
});
it('preserves the compact session-file summary', () => {
const summary =
'Session files: 4 persisted file(s) are available in /mnt/data, including 1 image(s). ' +
'Use known /mnt/data paths directly in later code-tool calls. ' +
'The app displays files/images automatically; do not invent download links or wrap generated images in Markdown.';
const input = ['stdout:', 'Report generated', '', 'Generated files:', summary].join('\n');
const output = cleanCodeToolOutput(input);
expect(output).toBe(input);
});
it('passes through output that contains no boilerplate (no false positives)', () => {
const input = 'stdout:\nHello, world!\n';
expect(cleanCodeToolOutput(input)).toBe('stdout:\nHello, world!');
});
it('does not strip the word "Note:" mid-output (only trailing paragraphs)', () => {
// A user-authored echo `Note: this is real content` mid-output should
// not be eaten by the trailing-Notes regex; only paragraphs that
// start with `Note:` AFTER a blank line near end-of-string are matched.
const input = [
'stdout:',
'Note: this is real content',
'more output',
'',
'Generated files:',
'- /mnt/data/foo.txt',
].join('\n');
const output = cleanCodeToolOutput(input);
expect(output).toMatch(/Note: this is real content/);
});
it('preserves a user-authored "Note:" line that follows a blank line in stdout', () => {
/** Tightened anchor: the trailing-notes regex now requires one of the
* known boilerplate openings (`Files from previous executions`, etc.)
* after `Note:`. A user `print()` that happens to output a blank
* line followed by `Note: ...something else...` must survive
* untouched, along with everything that follows it. */
const input = [
'stdout:',
'first line',
'',
"Note: this is the user's measurement output",
'more output',
'',
'Generated files:',
'- /mnt/data/foo.txt',
].join('\n');
const output = cleanCodeToolOutput(input);
expect(output).toMatch(/Note: this is the user's measurement output/);
expect(output).toMatch(/Generated files:/);
expect(output).toMatch(/- \/mnt\/data\/foo\.txt$/);
});
it('drops trailing blank lines after stripping', () => {
const input =
'stdout:\nresult\n\n\nNote: Files from previous executions are automatically available and can be modified.';
expect(cleanCodeToolOutput(input)).toBe('stdout:\nresult');
});
it('does NOT mutate stdout that legitimately contains an annotation phrase outside a file-list section', () => {
/** Codex regression: the previous global `.replace` would strip
* any line ending in one of the three annotation phrases — even
* if that line was the user's own stdout. A script doing
* `echo "foo | File is already downloaded by the user"` had its
* output silently mangled before being fed back into model
* context. With section-scoped stripping, the script's stdout
* passes through unchanged because there's no `Generated files:`
* or `Available files (...)` header above it. */
const input = [
'stdout:',
'foo | File is already downloaded by the user',
'bar | Image is already displayed to the user',
'baz | Available as an input — already known to the user',
].join('\n');
const output = cleanCodeToolOutput(input);
expect(output).toMatch(/foo \| File is already downloaded by the user/);
expect(output).toMatch(/bar \| Image is already displayed to the user/);
expect(output).toMatch(/baz \| Available as an input — already known to the user/);
});
it('strips annotations inside a file-list section but preserves identical phrases in stdout above it', () => {
/** Mixed case: the same execution's stdout coincidentally echoes
* one of the annotation phrases AND a file gets generated. The
* stdout phrase must survive; the file-listing annotation must
* still get stripped. */
const input = [
'stdout:',
'echo output: foo | File is already downloaded by the user',
'',
'Generated files:',
'- /mnt/data/foo.txt | File is already downloaded by the user',
].join('\n');
const output = cleanCodeToolOutput(input);
// Stdout line preserved verbatim.
expect(output).toMatch(/echo output: foo \| File is already downloaded by the user/);
// File-listing line had its annotation stripped.
expect(output).toMatch(/- \/mnt\/data\/foo\.txt$/m);
// The annotation phrase appears EXACTLY once now (only in stdout,
// not on the file-listing line).
const phraseCount = (output.match(/File is already downloaded by the user/g) ?? []).length;
expect(phraseCount).toBe(1);
});
});