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
🗂️ refactor: Clarify Code Sandbox File Guidance (#13236)
Danny Avilab9d4a595b93b
diff --git a/packages/api/src/agents/cleanup.spec.ts b/packages/api/src/agents/cleanup.spec.tsindex 077007de3b..563b5a67d3 100644--- a/packages/api/src/agents/cleanup.spec.ts+++ b/packages/api/src/agents/cleanup.spec.ts@@ -77,6 +77,22 @@ describe('cleanCodeToolOutput', () => {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!');diff --git a/packages/api/src/agents/cleanup.ts b/packages/api/src/agents/cleanup.tsindex 9025d2ea7c..ffb051901c 100644--- a/packages/api/src/agents/cleanup.ts+++ b/packages/api/src/agents/cleanup.ts@@ -3,8 +3,8 @@* tool output before LibreChat re-injects it into the assistant's* conversation history.*- * The bash executor in `@librechat/agents` appends two kinds of noise- * to every successful run:+ * 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@@ -21,12 +21,14 @@* per-file annotations are redundant *and* phrased inconsistently* ("downloaded" vs. "displayed" vs. "known to the user").*- * 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.+ * 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.*//**diff --git a/packages/api/src/agents/handlers.ts b/packages/api/src/agents/handlers.tsindex 4690cf73e4..51fed861be 100644--- a/packages/api/src/agents/handlers.ts+++ b/packages/api/src/agents/handlers.ts@@ -1179,11 +1179,11 @@ export function createToolExecuteHandler(options: ToolExecuteOptions): EventHand* call. Almost always means the seeding chain* (primeCodeFiles → initialSessions →* CodeSessionContext) dropped the file upstream.- * `session_id` is still emitted; agents falls- * through to the `/files/<sid>` legacy fetch- * which is post-cutover broken (returns 400).- * Pair with `[primeCodeFiles]` traces below to- * locate the layer that lost the ref. */+ * `session_id` is still emitted for continuity, but+ * concrete file refs must arrive through+ * `_injected_files`; agents no longer falls back to+ * `/files/<sid>`. Pair with `[primeCodeFiles]`+ * traces below to locate the layer that lost the ref. */logger.warn(`[code-env:inject] tool=${tc.name} _injected_files=0 — sandbox will see no input files`,{diff --git a/packages/api/src/agents/tools.spec.ts b/packages/api/src/agents/tools.spec.tsindex 2ad9c30db9..0643259847 100644--- a/packages/api/src/agents/tools.spec.ts+++ b/packages/api/src/agents/tools.spec.ts@@ -211,8 +211,10 @@ describe('registerCodeExecutionTools', () => {const readFile = result.toolDefinitions.find((d) => d.name === 'read_file');expect(readFile?.description).toContain('code-execution sandbox');expect(readFile?.description).toContain('/mnt/data/');+ expect(readFile?.description).toContain('Do not run ls/find');+ expect(readFile?.description).toContain('/tmp is per-call scratch');expect(readFile?.description).toContain('truncated around 256KB');- expect(readFile?.description).toContain('may return an error');+ expect(readFile?.description).toContain('true filesystem discovery');expect(readFile?.description).not.toContain('{skillName}');expect(readFile?.description).not.toContain('SKILL.md');expect(JSON.stringify(readFile?.parameters)).not.toContain('{skillName}');diff --git a/packages/api/src/agents/tools.ts b/packages/api/src/agents/tools.tsindex 268b7df01a..59c7ade5d8 100644--- a/packages/api/src/agents/tools.ts+++ b/packages/api/src/agents/tools.ts@@ -100,17 +100,9 @@ const READ_FILE_DEF: LCTool = Object.freeze({responseFormat: ReadFileToolDefinition.responseFormat,}) as LCTool;-const CODE_READ_FILE_DESCRIPTION = `Read the contents of a file from the code-execution sandbox or from prior code-execution output. Returns text content with line numbers for easy reference.+const CODE_READ_FILE_DESCRIPTION = `Read a known text file from the code-execution sandbox. Returns line-numbered text; large files may be truncated around 256KB.-BEHAVIOR:-- Text files: returned with numbered lines.-- Large text files are truncated around 256KB with a note to use bash_tool for the full content.-- Binary files and formats that are not safe to serialize as text may return an error. Use bash_tool to inspect or process them.--CONSTRAINTS:-- Only files produced by code execution or attached to the code-execution sandbox are accessible.-- Use paths returned by tool output or paths under /mnt/data/.-- Do not guess file paths. Use bash_tool to inspect available sandbox files when needed.`;+Use for text, CSV, JSON, Markdown, logs, and small source files at paths returned by tool output, just written, or under /mnt/data/. Do not run ls/find just to rediscover known paths. Use bash_tool for binary files, large files, transforms, metadata, or true filesystem discovery. /tmp is per-call scratch and unavailable later.`;const CODE_READ_FILE_PARAMETERS: LCTool['parameters'] = Object.freeze({type: 'object',