- Fix tool name lost on ToolResult events (persist via tool_name_by_call)
- Fix tool input lost on ToolResult events (persist via tool_args_by_call)
- Fix tool output in wrong field (error -> output)
- Fix text doubling in streaming (defer emit to ItemCompleted)
- Fix missing delta field in text streaming events
- Default server mode to no-token when --token not specified
- Add install-fast-sa and install-fast-gigacode justfile targets
## Summary
- Adds a `session_created` telemetry event that fires when a new session is created
- Logs safe session config fields: agent, agent_mode, permission_mode, model, variant
- Does not include workdir or any sensitive paths
- Respects existing telemetry enabled/disabled flag (`--no-telemetry`)
- Sends asynchronously via spawned task to avoid blocking session creation
## Test plan
- [x] `cargo check` passes
- [ ] Verify event reaches telemetry endpoint with correct fields
- [ ] Verify no event is sent when `--no-telemetry` is set
* fix: add postinstall chmod for npm binary permissions
* fix: report npm package version instead of compiled binary version
The --version flag now reports the version from package.json instead of the
version compiled into the Rust binary. This ensures the version matches what
was installed via npm, even when binaries are reused from previous releases.
* fix: bake version into binary at build time
Instead of hacking around the version in the Node.js wrapper script,
properly pass the version at build time via SANDBOX_AGENT_VERSION env var.
Changes:
- build.rs: Generate version.rs with VERSION constant from env var
- main.rs: Use generated version constant for clap --version
- Dockerfiles: Accept SANDBOX_AGENT_VERSION as build arg
- build.sh: Pass version as second argument to Docker builds
- release.yaml: Pass version to build script during CI
- Remove version hack from sdks/cli/bin/sandbox-agent wrapper
The version is now baked into the binary during the release build,
ensuring --version reports the correct npm package version.
* chore: remove inspect.sandboxagent.dev in favor of /ui/
* chore: add 404 page
* fix: correct inspector package name in Dockerfiles and add .dockerignore
- Change @anthropic-ai/sdk-inspector to @sandbox-agent/inspector in all Dockerfiles
- Add .dockerignore to exclude target/, node_modules/, etc from Docker context
The wrong package name caused pnpm install --filter to match nothing, so the
inspector frontend was never built, resulting in binaries without the /ui/ endpoint.
* chore: cargo fmt
* chore(release): update version to 0.1.4-rc.7
* feat: add Claude adapter improvements for HITL support
- Add question and permission handling for Claude sessions
- Add Claude sender channel for interactive communication
- Add stream event and control request handling
- Update agent compatibility documentation
* fix: restore Claude HITL streaming input and permission handling
- Add streaming_input field to SpawnOptions for Claude stdin streaming
- Enable --input-format stream-json, --permission-prompt-tool stdio flags
- Pipe stdin for Claude (not just Codex) in spawn_streaming
- Update Claude capabilities: permissions, questions, tool_calls, tool_results, streaming_deltas
- Fix permission mode normalization to respect user's choice instead of forcing bypass
- Add acceptEdits permission mode support
- Add libc dependency for is_running_as_root check
- Enable CORS for https://inspect.sandboxagent.dev by default
- Add --no-inspector-cors flag to opt out
- Additional --cors-allow-origin flags are now cumulative with inspector
- Inspector now tries current origin first before localhost:2468 fallback
Previously used cfg!(target_env = "musl") which checks compile-time,
causing musl-compiled sandbox-agent to always download musl binaries
even on glibc systems like Debian/E2B.
Now checks for /lib/ld-musl-*.so.1 at runtime to detect the actual
system libc and download the correct Claude binary variant.
- Restructure examples into individual packages per provider (daytona, docker, e2b, vercel) with shared utilities in @sandbox-agent/example-shared
- Make Claude plan mode prompt-only (no longer requires permissionMode=plan)
- Claude now defaults to bypass permission mode
- Add agent_file_edit_flow test for file editing capabilities
- Fix Daytona file permission setting to use executeCommand
