ragleap-agents

version — loading live from PyPI…

A small agent loop over ragleap-tools Tool objects. The model proposes one action at a time and you bring it as a plain callable; the library has no provider code and no database.

pip install ragleap-agents

Highlights

• Every proposal is checked against the tool's JSON Schema; an invalid proposal stops the run and no tool executes
• Tool results are untrusted: capped, fenced in <observation> tags, with lookalike tags neutralised case-insensitively
• Taint rule: after a tool declared taints=True has run, every later outbound=True tool needs approval. A tool with no declared policy counts as both, so it fails closed
• Pause and resume through a pluggable StateStore; a replayed or mismatched approval raises ResumeError

Tech stack

Python 3.10 to 3.12. One dependency: ragleap-tools. No provider code, no database.

Testing

29 test functions across 2 test files, verified directly from the repo (parametrised cases count once). CI runs 41 test cases on Python 3.10, 3.11 and 3.12, with a scripted model and no network.

Verification status

• Verified: the 41 scripted-model tests; mutation checks (removing the taint rule, the fence, the exception-text rule, the resume claim, the hard cap, the fail-closed default or the bool-is-not-integer check each fails at least one test); the published 0.1.0 wheel installs on 3.10, 3.11 and 3.12 and passes a scripted smoke test.
• Not verified: any real model (no provider has been run through this loop yet), long runs, concurrent resumes across processes, and equivalence with RagLeap Core's agent loop.

Limits to know about

• It does not stop prompt injection. It limits what an injected instruction can do, and only as far as your ToolPolicy declarations are accurate.
• Tool descriptions (capped at 300 characters) are shown to the model, so tools from an untrusted source can inject through them.
• Single agent only in v0.1.0; multi-agent crews are planned, not built.

Quickstart

from ragleap_tools import CALCULATOR_TOOL
from ragleap_agents import Agent, Policy, TRUSTED

def my_llm(prompt: str) -> str:
    ...  # call any model, return its text

agent = Agent(
    llm=my_llm,
    tools=[CALCULATOR_TOOL],
    policy=Policy(max_steps=4, tools={"calculator": TRUSTED}),
)
result = agent.run("What is 6 * 7?")
print(result.status, result.answer)