Tutorials¶
Short, ordered walks. Each builds on Quick Start.
1. First governed file operation¶
No token required.
runtime bootstrap
runtime files write ./tutorial.txt "hello from runtime"
runtime files read ./tutorial.txt
runtime audit tail -n 5
You used a provider operation. Policy, context and audit still ran.
2. First GitHub operation¶
export RUNTIME_GITHUB_TOKEN=ghp_…
runtime auth status
runtime github user get
runtime github repo view cli/cli
3. Watch policy deny something¶
The Command Engine refuses before git runs. The denial is audited like a
success would be. Governance shape:
Policy.
4. Run a shipped capability¶
runtime capability validate files/notes-roundtrip
runtime capability execute files/notes-roundtrip \
--input path=./cap.txt --input message="tutorial"
5. Write your own capability¶
Create ~/.engineering-runtime/capabilities/tutorial-echo.md:
# Tutorial echo
```runtime
version: v1
inputs:
path:
description: File to write
required: true
message:
description: Contents
required: true
workflow:
- provider: files
args: [write, "${path}", "${message}"]
- provider: files
args: [read, "${path}"]
```
runtime capability validate tutorial-echo
runtime capability execute tutorial-echo \
--input path=./mine.txt --input message=hello
That's the mechanics. For a capability built around a real task — including how to find the right operations and prove them before you write — see Create a Capability. Full grammar: Authoring reference.
6. Point at a shared capability directory¶
export RUNTIME_CAPABILITIES_DIR=~/work/engineering-runtime-capabilities
runtime bootstrap
runtime capability validate github/github-repo-health
Sharing a Capability Directory.
7. Same work in CI¶
Copy the minimal workflow into a repo that has
RUNTIME_GITHUB_TOKEN, or clone
engineering-runtime-samples
and run runtime-ci via workflow_dispatch.
8. Same work via an AI agent¶
Open
engineering-runtime-ai-agent
with runtime on PATH, then ask: "Run the files notes-roundtrip
capability into ./agent.txt". The agent should only invoke runtime.
Details: AI agent.