pi-workerlisted

# Pi Worker Keep product, architecture, scope, and integration decisions in the parent agent. Delegate only bounded execution. Never ask a worker to delegate again. Use the fewest useful workers. There is no fixed worker limit. Run tasks concurrently only when their writes and responsibilities are disjoint; honor any worker count or budget set by the user. Require a trusted workspace. Pi tools, including `bash`, run with the user's host permissions. Confirm `pi` is available without inspecting credentials. If Pi or provider access is unavailable, stop and give the relevant setup step. Create a private temporary directory and one Markdown task file per worker. Run each worker from the target workspace: ```sh pi -p \ --approve \ --no-session \ --no-extensions \ --no-skills \ --no-prompt-templates \ --no-themes \ --tools <task-specific-tools> \ --model <exact-provider/model> \ --thinking <level> \ @/absolute/path/to/task.md ``` Use `read,grep,find,ls` for read-only work. Add `edit,write,bash` only when the task requires workspace changes or commands. Omit `--model` or `--thinking` when the user did not specify it, allowing Pi's configured default. Preserve every explicit value unchanged. Never choose or silently substitute a model or effort level. Use the parent agent's native process control to run independent workers in parallel. Apply a bounded timeout and capture stdout, stderr, and exit status separately. Wait for every started worker; on cancell