Zamil Majdy 29da8db48e feat(copilot): E2B cloud sandbox — unified file tools, persistent execution, output truncation (#12212) ...

## Summary

- **E2B file tools**: New MCP tools
(`read_file`/`write_file`/`edit_file`/`glob`/`grep`) that operate
directly on the E2B sandbox filesystem (`/home/user`). When E2B is
active, these replace SDK built-in `Read/Write/Edit/Glob/Grep` so all
tools share a single coherent filesystem with `bash_exec` — no sync
needed.
- **E2B sandbox lifecycle**: New `e2b_sandbox.py` manages sandbox
creation and reconnection via Redis, with stale-key cleanup on
reconnection failure.
- **E2B enabled by default**: `use_e2b_sandbox` defaults to `True`; set
`CHAT_USE_E2B_SANDBOX=false` to disable.
- **Centralized output truncation**: All MCP tool outputs are truncated
via `_truncating` wrapper and stashed (`_pending_tool_outputs`) to
bypass SDK's head-truncation for the frontend.
- **Frontend tool display**: `GenericTool.tsx` now renders bash
stdout/stderr, file content, edit diffs (old/new), todo lists, and
glob/grep results with category-specific icons and status text.
- **Workspace file tools + E2B**: `read_workspace_file`'s `save_to_path`
and `write_workspace_file`'s `source_path` route to E2B sandbox when
active.

## Files changed

| Area | Files | What |
|------|-------|------|
| E2B file tools | `sdk/e2b_file_tools.py`, `sdk/e2b_file_tools_test.py`
| MCP file tool handlers + tests |
| E2B sandbox | `tools/e2b_sandbox.py` | Sandbox lifecycle
(create/reconnect/Redis) |
| Tool adapter | `sdk/tool_adapter.py` | MCP server, truncation, stash,
path validation |
| Service | `sdk/service.py` | E2B integration, prompt supplements |
| Security | `sdk/security_hooks.py`, `sdk/security_hooks_test.py` |
Path validation for E2B mode |
| Bash exec | `tools/bash_exec.py` | E2B execution path |
| Workspace files | `tools/workspace_files.py`,
`tools/workspace_files_test.py` | E2B-aware save/source paths |
| Config | `copilot/config.py` | E2B config fields (default on) |
| Truncation | `util/truncate.py` | Middle-out truncation fix |
| Frontend | `GenericTool.tsx` | Tool-specific display rendering |

## Test plan

- [x] `security_hooks_test.py` — 43 tests (path validation, tool access,
deny messages)
- [x] `e2b_file_tools_test.py` — 19 tests (path resolution, local read
safety)
- [x] `workspace_files_test.py` — 17 tests (ephemeral path validation)
- [x] CI green (backend 3.11/3.12/3.13, lint, types, e2e)

2026-03-03 21:31:38 +00:00

37 B

Raw Permalink Blame History

View Raw