v0.3.1

0.3.1 — first public release https://aigenos.io

The first version published to PyPI and npm. Earlier version numbers were internal builds; the two notes in the docs about names that changed "since the pre-release builds" refer to those.

What is in it:

The engine (pip install aigenos-flow): a skill is a SKILL.md whose steps: are a list of five kinds of step (ask, say, do, confirm, chat), with when: for a condition, check: for what an answer must satisfy and repeat: for a list of items: 20 words in all, on one page. No schema block and no Python in the skill. Types, lookups and who may give an answer are worked out from the steps and from the tools' signatures.
Tools are plain functions in one shared folder (skills/tools/). A function marked @readonly is a lookup; any other needs a confirm: above it on every path that reaches it, runs at most once per flow run and is audited. Tools can also come from your code (tools=), from one skill's own scripts.py, or from an MCP server (aigenos_flow.mcp_tools.MCPTools, started locally or reached by URL). Tools loaded with LangChain's MCP adapters and async def tools work as they are, and save_request is built in.
A fail-closed loader. A misspelled word, a tool that changes something with no yes above it, an input nothing asks for, a condition that compares text with a number: each is a load error with a sentence that says what to change, before any conversation starts.
What you approve is what runs. A confirmation that releases a tool is a card built by the engine: the tool and its inputs. They are compared again when the tool is called; a difference means nothing runs and the yes is asked for again.
A yes from someone else. by: manager on a confirm: files a request and waits. That yes cannot arrive through the conversation, from the user, the model or a posted form.
A skill reads back as a procedure. python -m aigenos_flow.procedure skills/ prints each step in plain words, every way through the flow, and what must be true before each tool that changes something. No model is involved.
The runtime: replay from the top on every turn, so a corrected answer applies every rule below it again; at-most-once writes; a flow hands off to another skill with do: other-skill. A cancelled flow is ended, not forgotten: what it collected is on record for the model to answer from, "continue" picks it up without its code or its yes, and "start over" and "let me pick a different one" each take one message. wrap_agent and create_agent for LangGraph.
The long form (fields: / actions: / flow:, nine step kinds) is what steps: expands to, and stays available for free-form chat inside a flow and per-argument bindings. A Python @flow is the escape hatch for loops.
The widget server (pip install "aigenos-flow[server]"): FastAPI routers that stream typed widgets over AG-UI, A2UI and MCP-UI, and the backend the Studio runs against.
The React renderer (npm install @aigenos/flow): , useFlowChat, and 16 primitives.
The scaffold (npx aigenos-agentic-app my-app).
The Studio, a local single-user authoring tool that runs from a clone: a canvas where a skill is a column of steps checked by the loader as you edit (ten things to add, from a question to a repeat, someone else's yes or a chat), a Gallery of three templates written as steps, a Tools page, the Procedure view, an Approvals inbox, a widget designer, and a Playground that runs any skill over AG-UI, A2UI or MCP-UI and shows every frame it sends. A skill can be created from a sentence and changed from one; in both cases the model writes and the loader decides whether the result is valid.