pw code

pw code [flags] [model]

  pw code
  
  # Start with a specific model (see 'pw ai models ls')
  pw code Jake.Thayne:openai/gpt-5.2
  
  # One-shot prompt
  pw code -p "find all TODO comments" Jake.Thayne:openai/gpt-5.2
  
  # JSON output for automation
  pw code -p "list Go files" -o json Jake.Thayne:openai/gpt-5.2
  
  # Resume the most recent session on its own model
  pw code -r latest
  
  # Pick a session to resume from a list
  pw code --resume
  
  # Read-only mode
  pw code --permission-mode read-only Jake.Thayne:openai/gpt-5.2
  
  # Use an org provider with a budget allocation
  pw code --allocation PW-QA-TEST org:openai/gpt-5.2
  
  # Think harder on every turn
  pw code --effort high Jake.Thayne:openai/gpt-5.2
  
  # Let the agent edit a directory outside the workspace
  pw code --add-dir /data/shared Jake.Thayne:openai/gpt-5.2
  
  # Run a custom agent as the main session
  pw code --agent reviewer
  
  # Pair with a notebook in your platform workspace and run code in its kernel
  pw code --notebook workspace:analysis/eda.ipynb --kernel my-kernel
  
  # Make this session viewable and drivable from the web app
  pw code --remote

      --add-dir strings                Directories outside the workspace the agent may read, search, and edit without prompting (repeatable)
      --agent string                   Run a custom agent definition as the main session (a name from .agents/agents, ~/.config/agents/agents, or --agents)
      --agents string                  Ad-hoc agent definitions as JSON keyed by name, merged over the definition files, e.g. '{"reviewer":{"description":"Reviews diffs","prompt":"...","tools":["ReadFile","GrepSearch"]}}'
      --allocation string              Budget allocation name (required for org provider models)
      --allowedTools strings           Tools pre-approved without prompting, e.g. ReadFile, 'Bash(git status *)', or 'ReadFile(/data/**)' to allow reads outside the workspace (repeatable; quote rules that contain commas)
      --dangerously-skip-permissions   Start in bypass-permissions mode and keep it in the shift+tab cycle; every tool call runs without asking
      --effort string                  Reasoning effort: default, none, minimal, low, medium, high, xhigh, or max (support varies by model)
  -h, --help                           help for code
      --kernel string                  The platform kernel the paired notebook runs on: the agent can run code in it (needs --notebook)
  -m, --model string                   Model ID, e.g. Jake.Thayne:openai/gpt-5.2 or org:openai/gpt-5.2 (run 'pw ai models ls' to list)
      --no-tools                       Disable tool calling (chat-only mode)
      --notebook string                Pair with this notebook (.ipynb), a local path or workspace:<path in your workspace> from any machine: the agent can read and edit its cells
  -o, --output string                  Output format for -p runs: text (the answer on stdout, progress on stderr) or json (one result object) (default "text")
      --permission-mode string         Permission mode: read-only, accept-edits, auto, bypass-permissions, plan (default: the saved mode, else accept-edits)
  -p, --prompt string                  Run one prompt non-interactively and exit; approvals are auto-denied, and the saved model and allocation apply when none is given
      --remote                         Let this session be viewed and driven from the web app's Agents page (the same as /rc); --remote=false overrides the remoteControl.newSessions setting
  -r, --resume string[="pick"]         Resume a session on this machine: pass a session ID or 'latest', or use --resume alone to pick from a list. The session keeps its model unless one is given
  -w, --workspace string               Workspace root directory (default: current directory)

      --context string         The context to use. Overrides PW_CONTEXT environment variable and current context from config.
      --platform-host string   The Parallel Works ACTIVATE platform host to use. Overrides PW_PLATFORM_HOST environment variable and the current context's server.
  -v, --verbose                Enable verbose logging