using-oi

name	using-oi
description	Use when an Agent needs to load, validate, interpret, or directly execute a versioned .oi program or Oi Skill.

Using Oi

1. Logical-read nearest oi.mod and exact installed execution.md once each; never default/combine/reinterpret or use packaging version.
2. Close exact-identity project/runtime/std graph in std byte order; validate entry/effect mapping. Compatibility probes are separate admission attempts, never target operations.
3. Exact 0.0.1 binds ordered typed inputs, maps reachable effects/authorities, host-invokes main, and exposes only mapped concrete typed Reply. This released route has no sealed admission, CONTEXT_ISOLATION_UNAVAILABLE, journal, or receipt.
4. For exact 0.0.2, bind inputs and admit only sealed verified artifacts, typed inputs, exact mappings/authorities/policy, trace/checkpoints, and protected journal.
5. Exclude conversation, expected output, repository prose, memory, prior claims/tool output, host objects, unmapped values, and capabilities.
6. Unavailable isolation stops 0.0.2 pre-entry: CONTEXT_ISOLATION_UNAVAILABLE, failed receipt, Policy none, zero operations; else a fresh Agent host-invokes main.
7. Deliver mapped typed Reply unchanged, receipt separately; canonical values stay only in the protected journal, unreadable by target.

Unsupported bootstrap has no receipt. Never Oi-call main, invent Reply/receipt fields, infer expected payloads, or expose ambient context. Supports 0.0.1/0.0.2.