agent-shell-py 0.1.8

A lightweight abstraction for executing CLI coding agents headlessly

Agent Shell

Agent Shell is a light weight abstraction for executing a cli coding agent headlessly and returning the output that can be used programatically as a unified contract

Installation

uv add agent-shell-py

or with pip:

pip install agent-shell-py

Examples

Execute

from agent_shell.shell import AgentShell
from agent_shell.models.agent import AgentType

shell = AgentShell(agent_type=AgentType.CLAUDE_CODE)

response = await shell.execute(
    cwd="/path/to/project",
    prompt="Can you tell me about this project?",
    allowed_tools=["Read", "Glob", "Grep"],
    model="sonnet",
)

print(response.response)
print(f"Cost: ${response.cost:.4f}")
print(f"Session: {response.session_id}")

# Resume the conversation using the session_id
follow_up = await shell.execute(
    cwd="/path/to/project",
    prompt="Now refactor the auth module based on your findings",
    allowed_tools=["Read", "Edit", "Bash"],
    model="sonnet",
    session_id=response.session_id,
)

Stream

from agent_shell.shell import AgentShell
from agent_shell.models.agent import AgentType

shell = AgentShell(agent_type=AgentType.CLAUDE_CODE)

async for event in shell.stream(
    cwd="/path/to/project",
    prompt="Refactor the auth module",
    allowed_tools=["Read", "Edit", "Bash"],
    model="sonnet",
    effort="high",
    include_thinking=True,
):
    if event.type == "system":
        print(f"Session: {event.session_id}")
    else:
        print(f"[{event.type}] {event.content}")

OpenCode

from agent_shell.shell import AgentShell
from agent_shell.models.agent import AgentType

shell = AgentShell(agent_type=AgentType.OPENCODE)

response = await shell.execute(
    cwd="/path/to/project",
    prompt="Can you tell me about this project?",
    model="anthropic/claude-sonnet-4-5",
)

print(response.response)
print(f"Session: {response.session_id}")

# Resume the conversation using the session_id
follow_up = await shell.execute(
    cwd="/path/to/project",
    prompt="Now refactor the auth module based on your findings",
    model="anthropic/claude-sonnet-4-5",
    session_id=response.session_id,
)

Note: OpenCode's run mode auto-approves all tools. The allowed_tools and effort parameters are configured via opencode.json, not CLI flags.

Logging

Agent Shell uses Python's standard logging module. Configure the agent_shell logger to capture tool calls, session IDs, costs, and errors:

import logging

logging.getLogger("agent_shell").setLevel(logging.INFO)
logging.getLogger("agent_shell").addHandler(logging.StreamHandler())

Set to DEBUG for raw JSON events and full command arguments.

Copilot CLI

from agent_shell.shell import AgentShell
from agent_shell.models.agent import AgentType

shell = AgentShell(agent_type=AgentType.COPILOT_CLI)

response = await shell.execute(
    cwd="/path/to/project",
    prompt="Can you tell me about this project?",
    model="gpt-4o",
)

print(response.response)
print(f"Session: {response.session_id}")

# Resume the conversation using the session_id
follow_up = await shell.execute(
    cwd="/path/to/project",
    prompt="Now refactor the auth module based on your findings",
    session_id=response.session_id,
)

Note: Copilot CLI doesn't expose pricing data. The cost field on AgentResponse will always be 0.0. The duration field is populated from usage.totalApiDurationMs.

Supported CLI Agents:

• Claude Code
• OpenCode
• Copilot CLI
• Gemini CLI
• Codex