AI agent skill · by @sbhooley

AINL

Deterministic compiled graphs for ZeroClaw (importer, runtime, MCP tools)

Archived 2026-10-23 · Source updated 2026-09-03 · License: Apache-2.0. Check the original repository for the latest version.

SKILL.md instructions

# AINL for ZeroClaw

**AINL** is an AI-native language that compiles **deterministic graphs**. In ZeroClaw, use it to import workflows, compile them, run them, and expose **`ainl-mcp`** tools to the agent.

**Package:** `ainativelang` **v1.8.0** (this tree; **PyPI** after publish — **`docs/RELEASING.md`**). Optional extras: `[mcp]`, `[solana]`.

## Agent instructions

Follow this skill when the user is on **ZeroClaw** (or wants **deterministic `.ainl` graphs** with **import / compile / run / MCP**). Prefer **`./install.sh`** after `zeroclaw skills install …`, or the manual **`pip install 'ainativelang[mcp]' && ainl install-mcp --host zeroclaw`** path (alias **`ainl install-zeroclaw`**). After bootstrap, suggest the **morning briefing** prompt example below when it fits.

## When to use this skill

- The user wants **portable, compiled workflows** (`.ainl`) instead of ad-hoc prompts.
- They mention **import**, **compile**, **morning briefing**, **clawflows**, **ecosystem**, or **deterministic graphs**.
- They need **MCP** access to validation / compile / ecosystem listing (`ainl-mcp`).

## Install (pick one)

1. **From this skill directory (recommended with `zeroclaw skills install`):** run `./install.sh`  
   That upgrades `ainl[mcp]` and runs **`ainl install-mcp --host zeroclaw`** (pip self-upgrade, MCP registration hints, `~/.zeroclaw/bin/ainl-run` shim).

2. **Manual:**  
   `pip install 'ainativelang[mcp]' && ainl install-mcp --host zeroclaw`

### Restricted environments (OpenClaw, Clawbot, webchat sandboxes)

This skill is designed for environments with **PEP 668 externally-managed Python** (Python 3.13+, sudo blocked).

- Preferred no-root order is: `python -m venv` first, then `--user`, then `--break-system-packages` as a last resort.
- The installer automatically uses compatible fallback modes when default pip fails.
- Clean uninstall (if needed):

```bash
pip3 uninstall -y ainl mcp aiohttp langgraph temporalio
rm -rf /tmp/ainl-repo /data/.openclaw/workspace/skills/ainl /data/.local/lib/python3.13/site-packages/*ainl*
```

## Commands the user cares about

| Command | Purpose |
|--------|---------|
| `ainl import …` | Import Markdown or ecosystem sources into `.ainl` |
| `ainl compile …` | Compile / validate graphs |
| `ainl run …` | Execute a graph via the CLI runtime where applicable |
| `zeroclaw ainl run <file.ainl>` | After bootstrap: wrapper that compiles/runs through the installed shim |

## After install — prompt suggestion

Tell the user they can say in ZeroClaw:

> Import the morning briefing using AINL.

(They should point the importer at their Markdown or pack; `ainl import` subcommands match their source type.)

## MCP

Configure the host to run **`ainl-mcp`** as a stdio MCP server (see AINL docs: *External orchestration* / MCP). **`ainl install-mcp --host zeroclaw`** aligns local setup with ZeroClaw paths where applicable. Authoring tools include **`ainl_get_started`** (pass **`wizard_state_json`** to continue), **`ainl_step_examples`** (snippets by topic), validate/compile responses with **`contract_validation_status`**, and resources **`ainl://strict-valid-examples`** and **`ainl://strict-valid-families`**. See repo **`AGENTS.md`** and **`docs/operations/MCP_AINL_WIZARD_AND_CORPUS.md`**.

## References

- Upstream: [github.com/sbhooley/ainativelang](https://github.com/sbhooley/ainativelang)
- Package: **`ainativelang`** on PyPI