Metadata-Version: 2.5
Name: render-lab-tasks-agent
Version: 0.1.0
Summary: agent tasks for Render Workflows
Project-URL: Repository, https://github.com/render-lab/render-tasks-python
License-Expression: MIT
License-File: LICENSE
Requires-Python: >=3.12
Requires-Dist: httpx<0.29,>=0.28
Requires-Dist: render-lab-tasks-core<0.2,>=0.1.1
Requires-Dist: render-lab-tasks-llm<0.2,>=0.1.1
Requires-Dist: render==1.0.1
Description-Content-Type: text/markdown

# Agent tasks for Python

Eight registered tasks cover planning, single decisions, loops, reflection,
routing, history compression, named tools, and injected KV memory.
Import the tasks and `app` from `render_lab_tasks_agent.tasks`.

## Composition and environment

LLM calls inherit the `render-lab-tasks-llm` provider/model/ledger environment
contract. Credentials remain lazy. Raw decision implementations accept
`CompletionDeps(complete=...)`; loops take `AgentDeps(step=..., tools=...)`.
The default loop dispatches wrapped `agent.step` through `ctx.run` so every
model decision is a durable child run. Register `ToolBinding` values with
`register_agent_tool`; each binding's `run(ctx, args)` must dispatch wrapped
Render tasks via `ctx.run` and return a string observation.

`memory_impl` requires an injected `MemoryDeps(kv=...)`; there is no implicit
in-memory fallback or default KV account. Re-register a wrapper under your own
namespace to supply the store. No live agent/provider run verified this port.

## Installation

```sh
pip install render-lab-tasks-agent==0.1.0
```
