From 2ebef6f796af11347f5c9f2d46657608598ae7a2 Mon Sep 17 00:00:00 2001 From: "fineas-bot[bot]" <258147136+fineas-bot[bot]@users.noreply.github.com> Date: Sun, 28 Jun 2026 18:22:28 -0400 Subject: [PATCH] Trim generic sandbox startup prompt Rebuilt via GitHub createCommitOnBranch so the PR branch contains a verified GitHub-created commit. --- services/sandbox/SYSTEM_PROMPT.md | 24 ------------------------ 1 file changed, 24 deletions(-) diff --git a/services/sandbox/SYSTEM_PROMPT.md b/services/sandbox/SYSTEM_PROMPT.md index 1b689bc75..5793196dc 100644 --- a/services/sandbox/SYSTEM_PROMPT.md +++ b/services/sandbox/SYSTEM_PROMPT.md @@ -130,11 +130,6 @@ | vlogs tool_analytics --start 7d → tool usage stats | vlogs query 'level:error AND event:tool_call_completed' --limit 20 → raw LogsQL | -[Ethereum Mainnet RPC] -|When you need an Ethereum mainnet RPC endpoint and the user has not specified another provider, use the Reth-hosted mainnet endpoints: -| HTTP: https://ethereum.reth.rs/rpc -| WSS: wss://ethereum.reth.rs/ws -| [Common Tool CLIs] |NEVER call external APIs directly via curl unless you are downloading a file the prompt explicitly told you to fetch that way. |Use the relevant tool CLI instead — it routes through the sandbox proxy and only exposes tools your deployment allows. @@ -197,22 +192,3 @@ |Verifying only the underlying code, local file, or intermediate state is not enough when the user cares about the rendered artifact, discoverable name, live integration, or execution result. |If you cannot verify the exact surface because of missing access, missing runtime support, or a failed check, say the work is partially complete and lead with the specific unverified gap and blocker. |Do not say or imply that the task is done, fixed, working, or shipped when the exact user-visible surface remains unverified. - -[Document processing — built-in libraries] -|The sandbox has these Python libraries pre-installed for reading documents. -|Always invoke them via `uv run python` (per the [Python policy] above) — never `python3`. -| -|.docx files (python-docx): -| uv run python -c "from docx import Document; doc=Document('file.docx'); print('\n'.join(p.text for p in doc.paragraphs))" -| -|.xlsx files (openpyxl): -| uv run python -c "from openpyxl import load_workbook; wb=load_workbook('file.xlsx'); ws=wb.active; [print(row) for row in ws.iter_rows(values_only=True)]" -| -|.pptx files (python-pptx): -| uv run python -c "from pptx import Presentation; prs=Presentation('file.pptx'); [print(shape.text) for slide in prs.slides for shape in slide.shapes if shape.has_text_frame]" -| -|.pdf files (pymupdf): -| uv run python -c "import fitz; doc=fitz.open('file.pdf'); [print(page.get_text()) for page in doc]" -| -|For longer scripts, create a .py file and run it with `uv run path/to/script.py` instead of one-liners. -|ALWAYS use these libraries to extract text from documents — never try to parse raw XML or binary.