| name | execute_python |
| description | Execute Python code in an E2B Firecracker sandbox with human review. |
| type | tool |
| category | code |
| tool | {"module":"app.skills.execute_python.tool","function":"execute_python","async":true} |
| metadata | {"author":"oniva","version":"2.0.0"} |
| audit-level | high |
Execute Python
Execute Python code in an E2B Firecracker microVM sandbox.
Execution Modes
E2B Sandbox (ENABLE_SANDBOX=True + E2B_API_KEY set)
- Code runs in Firecracker microVM with hypervisor isolation
- Requires human review approval before execution
- Timeout: 60 seconds
- Full Python environment (pandas, numpy, etc.)
- Network isolated by Firecracker
- Output files saved to workspace/code_outputs/
Legacy In-Process (ENABLE_SANDBOX=False)
- Falls back to restricted in-process sandbox
- AST validation blocks dangerous patterns
- Timeout: 5 seconds (max: 30)
- Limited to safe builtins and allowed imports
Parameters
| Parameter | Type | Required | Default | Description |
|---|
code | string | Yes | - | Python code to execute |
input_files | list[str] | No | [] | Workspace files to upload to sandbox |
timeout | int | No | 60 | Max execution time in seconds |
Response (E2B Mode)
{
"success": true,
"status": "pending_review",
"ticket_id": "uuid",
"message": "Code submitted for human review."
}
Response (Legacy Mode)
{
"success": true,
"output": "6\n",
"result": "6",
"errors": null,
"error": null
}
Agent Guidance
- If code execution is rejected, modify the code to address the reviewer's concerns
- Do not retry the same code more than 2 times
- Use input_files to pass workspace data to the sandbox
- Output files are automatically saved to workspace/code_outputs/
Migration from v1.x
- Timeout changed: 5s (in-process) -> 60s (E2B remote)
- New parameter:
input_files for workspace file upload
- E2B mode returns
pending_review status instead of immediate results
- Legacy mode preserved as fallback when ENABLE_SANDBOX=False
