Refactor llm-judge.ts from raw API-only to dual-mode: CLI first
(spawns claude with --output-format text, unsets CLAUDECODE), API
fallback. Backend selectable via LLM_JUDGE_BACKEND env var.
Fix pre-existing race condition in config.rs tests where parallel
test execution caused env var mutations to interfere. Added static
Mutex to serialize env-mutating tests.
Add periodic PRAGMA wal_checkpoint(TRUNCATE) every 5 minutes for both
sessions.db and btmsg.db to prevent unbounded WAL growth under sustained
multi-agent load. Improve Landlock fallback log message with kernel
version requirement. Add WAL checkpoint tests.
SandboxConfig with RW/RO paths applied via pre_exec() in sidecar child
process. Requires kernel 6.2+ with graceful fallback. Per-project toggle
in SettingsTab. 9 unit tests.
Defense-in-depth: Claude CLI uses credentials file for auth, not
ANTHROPIC_API_KEY from env. OPENAI_* intentionally kept (Codex runner
needs it). 8 unit tests for strip_provider_env_var.
Add switcher-claude multi-account support with profile selector in AgentPane
toolbar, skill autocomplete menu (type / in prompt), and 5 new AgentQueryOptions
fields (setting_sources, system_prompt, model, claude_config_dir,
additional_directories) flowing through full stack from Rust to SDK.
New Tauri commands: claude_list_profiles, claude_list_skills, claude_read_skill,
pick_directory. New frontend adapter: claude-bridge.ts.
Add env_clear() + envs(clean_env) to Command in SidecarManager to
filter all CLAUDE-prefixed environment variables before spawning the
sidecar process. This provides primary defense against SDK nesting
detection when BTerminal is launched from a Claude Code terminal.
The JS-side stripping via SDK env option is retained as defense-in-depth.
Consolidated from two separate runners (agent-runner-deno.ts +
agent-runner.ts) to a single pre-built agent-runner.mjs that runs
under both Deno and Node.js. resolve_sidecar_command() now checks
runtime availability upfront before path search, with improved
error messages. Removed agent-runner-deno.ts from tauri.conf.json
bundled resources.
- Add permission_mode field to AgentQueryOptions (Rust, sidecar, bridge)
flowing from controller through sidecar to SDK; defaults to
bypassPermissions, supports default mode
- Fix AgentPane onDestroy bug: remove stopAgent() from onDestroy (fires
on layout remounts), move stop-on-close to TilingGrid onClose handler
- Bundle SDK into sidecar via esbuild (remove --external flag)
Claude CLI v2.1.69 hangs silently when spawned via child_process.spawn()
with piped stdio (known bug github.com/anthropics/claude-code/issues/6775).
Replace raw CLI spawning in both sidecar runners with the SDK's query()
function, which handles subprocess management internally. SDK message
format matches CLI stream-json, so the sdk-messages.ts adapter is
unchanged.
- agent-runner.ts: use SDK query() with AbortController for stop
- agent-runner-deno.ts: use npm:@anthropic-ai/claude-agent-sdk import
- sidecar.rs: add --allow-write and --allow-net Deno permissions
- package.json: add @anthropic-ai/claude-agent-sdk ^0.2.70, build:sidecar script
Create Cargo workspace at v2/ level with members: src-tauri,
bterminal-core, bterminal-relay. Extract PtyManager and SidecarManager
into shared bterminal-core crate with EventSink trait for abstracting
event emission. TauriEventSink wraps AppHandle. src-tauri pty.rs and
sidecar.rs become thin re-exports. Move Cargo.lock to workspace root.
Add v2/target/ to .gitignore.
