Start with AI¶
Goal: get Engineering Runtime installed and a first auth-free success by asking your assistant — without a GitHub token.
You do not need to learn Engineering Runtime to use it. You need to accomplish an engineering task. Your assistant reads these pages; the runtime executes, governed and audited.
Documentation is a product contract
Engineering Runtime documentation is an executable product contract for humans and agents: it exposes goals, inputs, commands, expected results and recovery. External tools and CI remain authoritative for effective context; provider platforms remain authoritative for credentials. Runtime resolves one safe context snapshot, applies policy, executes through the provider, and records that same snapshot in audit. Prose or model output cannot override those controls. File Engine, exact-executable and bounded-output limits are explicit; they are not a full host sandbox or universal DLP.
Your runtime carries its own contract
This site gets you installed, gets your first capabilities running, and
explains how the pieces fit. Any Runtime command prepares the Home
automatically; runtime bootstrap is the optional explicit report. Start at
<Runtime Home>/RUNTIME-AGENT.md (the contract) and manifest.json
(this binary's providers and operations). Specs and command cheatsheets
for your installed version live in <Runtime Home>/specs/ and
commands/. Read those for precise operational detail — they describe
the binary you actually have. Come back here for concepts, upgrades,
and anything not yet installed.
Paste this — minimal local path¶
Install Engineering Runtime by following
https://docs.engineeringruntime.com/get-started/installation/
then follow https://docs.engineeringruntime.com/get-started/quick-start/
exactly: bootstrap, confirm runtime capability list is an empty success,
run the auth-free files notes-roundtrip capability against a disposable
directory, and show me runtime audit tail. Do not ask for a GitHub token
yet. Do not use git, gh, curl, or /metadata/*.
Any capable assistant works. This prompt names URLs and a goal — not a vendor. Bring whichever assistant you already use.
More in the prompt library.
What it will do¶
- Install Runtime — one command, no GitHub account needed
- Bootstrap a clean Runtime Home
- Run the files path in a disposable directory
- Show the audit record
You supervise a named sequence rather than guessing what it did.
What it will ask you for¶
Things it cannot invent, and should never pretend to have:
| Input | Why |
|---|---|
Permission to write the binary (sometimes sudo) |
Installing to /usr/local/bin may need it |
| A GitHub token | Only for the authenticated follow-up, not this path |
If an assistant claims it finished the local path by calling gh or curl,
it bypassed Runtime. Check with the commands below.
How to verify it worked¶
runtime version prints the installed version.
runtime capability list on a fresh Home is empty and exits 0.
runtime config validate reports Config: compiled default and
Policy: compiled safety profile when you have not pointed at documents.
Authenticated GitHub follow-up¶
Only after the local path works. Paste this next:
Follow https://docs.engineeringruntime.com/authentication/github/
Ask me for a scoped RUNTIME_GITHUB_TOKEN and the owner/repo to use.
Then run runtime github user get and one github capability against that
repository. Do not invent an organisation. Do not use git, gh, or curl.
Going deeper¶
Capability authoring, recovery, and enterprise connection are follow-ups — see Create a Capability and CLI reference. Do not mix them into the first success.
For a later one-ask authoring task, tell the assistant the outcome and required
facts. It should run capability authoring-context, reuse an existing
definition where possible, write through runtime files, validate, plan, and
show you the source, diff, digest and plan. It must not publish or execute the
new capability unless you ask for that separately.
The Java-service showcase is not a first run: it needs a write-scoped token and creates real commits. Showcase: scaffold and ship a Java service.
What it will refuse to do¶
If you also apply the Runtime Agent contract — the
installed <Runtime Home>/RUNTIME-AGENT.md, not a website copy — the
assistant executes only through runtime — never raw gh, kubectl,
terraform or arbitrary shell.
That is a feature, not a limitation. It is what makes the run auditable.
Enforcement depends on your assistant
Some assistants (Claude Code, Cursor today) ship hook systems that can
mechanically block non-runtime shell. Most cannot — for those the
contract is a strong instruction the model chooses to follow, and nothing
stops it being ignored.
This is a statement about hook systems, not a recommendation of a vendor. See AI agent setup for what each can enforce.
What this page does not promise¶
Assistants are not deterministic. Two assistants, or the same one on different days, will not take identical steps. This page promises the goal chain and the verification commands — not exact assistant behaviour.
If it goes wrong, the commands above tell you where you actually are, and Troubleshooting covers the common failures.
Next¶
Operational examples on this site were verified against Runtime 0.9.8. After bootstrap, the version-exact files in Runtime Home win.