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
/**
* Strips repetitive boilerplate from `@librechat/agents` code-execution
* tool output before LibreChat re-injects it into the assistant's
* conversation history.
*
* Older bash executor output from `@librechat/agents` appended two
* kinds of noise to successful runs:
*
* 1. **Trailing "Note:" paragraphs** — long behavioral hints repeating
* rules the agent already has via its system prompt
* ("Files from previous executions are automatically available...",
* "Files in 'Available files' are inputs..."). Re-stating them on
* every tool call wastes tokens at scale (50+ tokens × N tool calls).
*
* 2. **Per-file annotations** — each file in the `Generated files:` /
* `Available files (...):` lists gets a `| <annotation>` suffix
* (`File is already downloaded by the user`,
* `Image is already displayed to the user`,
* `Available as an input — already known to the user`). The two
* section headers already convey the new-vs-known distinction; the
* per-file annotations are redundant *and* phrased inconsistently
* ("downloaded" vs. "displayed" vs. "known to the user").
*
* Newer compact summaries under `Generated files:` intentionally do
* not list `- /...` paths and pass through unchanged. Stripping happens
* in LibreChat (this file), not upstream, so the cleaning is reversible
* — pin to a specific upstream version and the patterns adjust here
* without releasing a new agents build. The patterns are anchored
* conservatively: only the documented forms are matched, so a future
* upstream string change leaves user-authored `Note:` lines (or
* legitimate `|`-delimited filenames) untouched.
*/
/**
* Matches a `Generated files:` or `Available files (...)` block —
* the section header line plus the file-listing lines that follow.
*
* Group 1 captures the header (with trailing newline). Group 2
* captures the file lines (each starts with `- /` per the executor's
* format, accommodating both line-per-file (≥ 4 files) and inline
* comma-separated (≤ 3 files: `- /a.txt | <ann>, - /b.txt | <ann>`)
* layouts because the latter is still a single line that begins with
* `- /`).
*
* The `Available files (...)` form has parenthesized content
* (`(inputs, not generated by this execution)`) which the
* `\([^)]*\)` segment matches without baking the exact phrasing in.
*
* Block ends at the first line that doesn't start with `- /` (next
* iteration of the `+` quantifier fails). The blank line preceding
* the trailing `Note:` paragraphs reliably terminates the block.
*/
const FILE_SECTION_PATTERN =
/((?:Generated files:|Available files \([^)]*\):)\n)((?:- \/[^\n]*(?:\n|$))+)/g;
/**
* Matches the per-file `| <annotation>` suffix the bash executor adds to
* each file inside a `Generated files:` / `Available files (...):` block.
* Two listing formats: line-per-file (≥ 4 files) and inline
* comma-separated (≤ 3 files: `- /mnt/a.txt | <ann>, - /mnt/b.txt | <ann>`).
* The non-greedy `[^,\n]*?` plus the `(?=,|\n|$)` boundary stops the match
* at either end-of-line or the inline `, ` separator so an inline list
* strips each file's annotation independently.
*
* Applied ONLY within file-section blocks captured by
* `FILE_SECTION_PATTERN`, never globally — a user script that legitimately
* echoes one of these phrases (`echo 'foo | File is already downloaded by
* the user'`) would otherwise have its stdout silently mutated.
*/
const PER_FILE_ANNOTATION_PATTERN =
/\s*\|\s*(?:File is already downloaded by the user|Image is already displayed to the user|Available as an input[^,\n]*?)(?=,|\n|$)/g;
/**
* Matches the bash executor's trailing `Note:` paragraphs. Anchored to
* the two known boilerplate openings (`Files from previous executions`,
* `Files in "Available files"`) so a user-authored `Note:` line — even
* one preceded by a blank line in stdout — is not eaten by the
* `[\s\S]*$` tail. New upstream variants need to be added here
* explicitly; that's the trade-off for not silently dropping content.
*/
const TRAILING_NOTES_PATTERN =
/\n\s*\n\s*Note:\s*(?:Files from previous executions|Files in "Available files")[\s\S]*$/;
/**
* Returns `content` with the bash-executor boilerplate removed. Safe
* to call on any tool output — non-matching text is returned unchanged.
*
* Annotation stripping is scoped: only the contents of `Generated
* files:` / `Available files (...):` blocks are mutated. Stdout that
* happens to contain a phrase like
* `| File is already downloaded by the user` (e.g. a script echoing
* the very string we're scrubbing) passes through unchanged.
*/
export function cleanCodeToolOutput(content: string): string {
if (!content) {
return content;
}
const noAnnotations = content.replace(
FILE_SECTION_PATTERN,
(_match, header: string, files: string) =>
header + files.replace(PER_FILE_ANNOTATION_PATTERN, ''),
);
const noTrailingNotes = noAnnotations.replace(TRAILING_NOTES_PATTERN, '');
return noTrailingNotes.trimEnd();
}