Yoke v0.6.2

Fixed

• CI test failures — runShellCommand now ensures the default workspace
  directory exists before spawning; tests no longer depend on the gitignored
  .test-data/ directory being present.

Yoke v0.5.0

Added

• CLI subcommands in the binary — the released executable can create
  users directly: yoke create-user <username> <password> (respects
  DATABASE_PATH). yoke version prints the version.

Yoke v0.4.0

Added

• YOLO mode — per-user 🔥 YOLO toggle (header button or /yolo on|off)
  disables approval prompts: shell commands, background jobs, and remote
  operations run immediately without confirmation. Requires explicit
  confirmation when enabling; POST /api/yolo to control via API.

Yoke v0.3.0

Added

• Stop actions — a ⏹ Stop button (or POST /api/agent/stop) cancels the
  running agent run; the run ends with a "Run stopped" notice instead of an
  error.
• Shell command timeouts — !timeout=<seconds> and timeout_seconds on
  POST /api/shell and the agent's run_command tool set a per-command limit
  (1-3600s, default 15s).
• Background shell jobs — !!<command> runs asynchronously: the request
  returns a job id immediately, the command waits for approval then runs in
  the background. New endpoints GET /api/shell/jobs,
  GET /api/shell/<id>, POST /api/shell/<id>/stop, a Shell jobs panel, and
  /shell, /shell stop, /shells commands.
• History restore — the conversation is restored into the view on login,
  page reload, and project switch (active + archived messages, tool calls and
  results included).
• Higher default iterations — default max agent iterations raised from 10
  to 100 (max 30,000).
• Configurable approval timeout — APPROVAL_TIMEOUT_MINUTES env var
  (default 5).

Fixed

• Approval prompts during agent runs — approval cards now render while the
  agent waits for permission on run_command/remote tools (polling previously
  only ran for ! shell commands, so prompts silently expired).
• Shell timeout/stop actually kills the command — Deno's AbortSignal
  does not kill subprocesses; output now goes to temp files and the process is
  terminated explicitly.

Yoke v0.2.0

Added

• Token streaming — the agent's reply now streams into the chat as it is
  generated, including a collapsible thinking block for reasoning content.
  Tool calls appear instantly and update as they complete.
• Thinking effort — per-user reasoning_effort selection
  (auto/low/medium/high) via a header dropdown, /thinking, or
  POST /api/thinking.
• Local shell commands — new run_command agent tool that executes
  commands in the workspace (approved from the web UI).
• Iteration indicator — the UI shows the current agent iteration
  (⚙️ Iteration N / M) during a run.
• Higher iteration limits — max agent iterations raised to 30,000.

Yoke v0.1.0

First release.

Added

• Agent harness — DeepSeek-powered coding agent with tools: read_file,
  write_file, list_directory, search_code, finish.
• Subagents — spawn background subagents (spawn_subagent), poll
  (check_subagent), or block for results (wait_subagent); live monitor in
  the UI.
• User accounts — username/password login with PBKDF2-hashed passwords and
  SQLite-backed session tokens. Users are created via the CLI
  (create-user.ts).
• Projects — each project has its own working directory and can pin its own
  LLM model; per-project conversation context.
• Context management — persistent per-user/per-project message history
  stored as append-only rows (tool calls and responses included). Compaction,
  summarization, and reset archive messages instead of deleting them;
  /history views them and /restore brings them back.
• Code & file indexing — incremental symbol extraction for 12 languages
  with search (/api/search, search_code tool).
• Web access — web_search (Bing RSS, no API key) and web_fetch
  (HTML → markdown) tools, plus /web and /fetch commands.
• Remote hosts — SSH inventory with key/password auth, remote_exec,
  remote_status, SFTP upload/fetch, and full directory deploys with
  post-deploy commands. sudo is supported via sudo -S with the password
  fed through stdin (never in command text).
• UI approvals — shell and remote commands require an in-UI permission
  prompt (Approve / Allow always / Deny) instead of a terminal prompt.
• Multiple models — per-user and per-project model selection
  (DEEPSEEK_MODELS, /model).
• Web UI — chat-style interface with dark mode, markdown rendering, hosts
  and subagents panels, project and model selectors, and slash commands
  (/help, /compact, /summarize, /usage, /history, /restore, …).
• Self-signed HTTPS — automatic certificate generation (ENABLE_HTTPS).
• Testing — Deno test suite (unit + integration against an in-process SSH
  server fixture): deno task test.

Security

• Passwords hashed with PBKDF2-SHA256 (100k iterations).
• Workspace sandboxing for file tools (path traversal protection).
• Shell execution gated by human approval in the UI.
• SSH/sudo passwords sent over the encrypted channel via stdin only.