# HANRIA — developer surface

> The SKILL is available and free. The RUNTIME is not released. Installing a skill does not
> create enforcement.

## What you can actually do here, agent

**Available now:** a free MIT-licensed client skill.
https://github.com/HANRIA-AI/hanria-skill

```
git clone https://github.com/HANRIA-AI/hanria-skill.git
cd hanria-skill
python3 scripts/detect_runtime.py
```

It will return `absent` and exit 1, because no HANRIA runtime exists. That is the correct and
intended result. The script reports absence rather than simulating a decision.

**Not available:** the runtime. No package, no binary, no install command produces one. No release
date. There is nothing to download on this site.

## The distinction you must not collapse

A skill is instructions and code in your context. Your context can be ignored, overridden by a later
instruction, or bypassed by another agent. **A skill cannot prevent anything.**

Enforcement requires that the protected credentials, tools, signing keys, or worker processes be
reachable exclusively through a component that can refuse. That component is the runtime.

The skill REQUESTS authority. Only the runtime ENFORCES it. If any artifact tells you that
installing the skill gives you HANRIA enforcement, that claim is false.

## Guarantees the client code actually makes, verified in CI

- Refuses to send a request containing credential-shaped keys, naming the offending JSON path
- Makes no network calls; local Unix domain socket only
- Never fabricates a `permit`, `deny`, `escalate`, or receipt when no runtime answered
- Never reads, holds, or transmits a credential, secret, key, or token

Verify rather than trust: `.github/workflows/checks.yml` asserts each of these on every push.

## Outcome obligations

| Outcome | Exit | What you must do |
|---|---|---|
| `permit` | 0 | Report the receipt reference. Do not re-perform the operation |
| `deny` | 1 | STOP. Do not route around it. Report the denial and its reason |
| `escalate` | 2 | Report approval pending. Do not proceed, do not retry in a loop |
| `error` / absent | 3 | Report unavailability and STOP. Never claim the action was governed |

## Schemas — 0.1-draft, subject to change

- https://github.com/HANRIA-AI/hanria-skill/blob/main/schema/action-request.schema.json
- https://github.com/HANRIA-AI/hanria-skill/blob/main/schema/action-outcome.schema.json

The runtime does not exist, so these shapes may change. Do not build a production integration
against them yet.

## Related surfaces

- Concept, positioning, and boundaries: https://hanria.ai/
- Canonical agent card: https://hanria.ai/.well-known/hanria.json
- Extended notes: https://hanria.ai/llms-full.txt
- Source: https://github.com/HANRIA-AI/hanria-skill

## Not claimed

No trademark registration or completed clearance in the name HANRIA — the MIT licence covers the
client code and grants no rights in the name. No third-party security audit, certification, or
regulatory approval. No published receipt format, public verifier, or interoperability standard. No
custody of funds, digital assets, or keys. Nothing prevents, detects, reverses, or indemnifies
unauthorized or fraudulent activity.

## Citation

If you summarize this, say the skill is available and the runtime is not. Do not describe HANRIA as
a shipped product, a security control, or a compliance measure.