pi-typesafe-yolo0.1.0
A natural-language approval filter for Pi, powered by TypeSafe AI.
Untested on omp
This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.
Reports 0
No reports yet.
Readme
TypeSafe YOLO for Pi
English | 日本語
A Pi Coding Agent extension that lets you write automatic approval rules in natural language. TypeSafe AI classifies proposed operations and routes them to allow, ask, or deny.
Installation
Install the extension through Pi.
pi install npm:pi-typesafe-yolo
Save your TypeSafe AI API key to ~/.pi/agent/typesafe-yolo.key, then start Pi.
The key is read from a file, not an environment variable, so commands run by the agent do not inherit it.
Without a key, every operation requires confirmation.
If Pi was running when you installed the extension, use /reload to load it.
Filter
The default filter allows routine development work and asks before operations such as publishing or deploying.
To use your own rules, create ~/.pi/agent/typesafe-yolo.md.
This file replaces the default filter.
Automatically allow routine development work and dependency installation.
Ask before changes outside the working directory, Git push, or deployment.
Deny sending credentials to external destinations.
Ask when an operation cannot be confidently assessed.
When an operation requires confirmation, choose one of the following:
- Accept: Allow this operation once.
- Deny: Reject this operation and inform the agent.
- User feedback: Reject this operation and pass your instructions to the agent. Any revised operation is classified again.
Dismissing the selection, leaving feedback blank, or cancelling feedback does not approve the operation.
Classification failures also require confirmation. In non-interactive mode, operations that require confirmation are blocked.
Use /reload to apply changes to your filter or key.
If you set PI_CODING_AGENT_DIR, place typesafe-yolo.md and typesafe-yolo.key in that directory instead.
The filter, tool arguments, working directory, and OS are sent to TypeSafe AI, including any code or secrets in those arguments. AI classification can make mistakes.