agent.py

This is to run different AI agents.

Installation

uv pip install git+https://github.com/trobz/agent.py

Usage

agent <instructions>

Run the above command to run the agent with instructions and default codex agent.

agent --instructions-file <instructions file name>
agent -f <instructions file name>

Alternatively run one of the above commands to run the agent with instructions in the file name.

agent --workflow <workflow file name>
agent -w <workflow file name>

Or one of the above command can be ran to run the agent with multiple instructions in workflow file name. The format of the workflow file is as follows:

- name: name of the work flow
  code: code of the work flow
  steps:
    - name: name of the step
      original: original instruction
      work_dir: optional work directory
      command: bash command to run
    - name: name of other step
      original: original instruction
      condition: optional bash command to run and check the return code
      instruction: instruction to run the agent
    - name: name of other step
      original: original instruction
      ignore: true
      notes: notes

step can have files set to a list of files to include in the instruction for the agent to consult.

step can have commit_if_change set to either true or a string to commit the changes. If set to true, the step name is used as the commit message. If set to a string, the string is used instead.

--backend <agent name>

Add the above option to run the agent with agent name. Available agents: codex, opencode, and gemini.

--mode <mode>

Add the above option to run the agent with the mode. Available modes: normal, yolo. Mode yolo is to run codex with --dangerously-bypass-approvals-and-sandbox option and run gemini with --yolo option.

--model

Add the above option to use a different model.