Skip to content
Back to skills

Oia Manifest

ASecurity

Emit .harness/oia-manifest.json declaring layer alignment with the OIA v0.1 9-layer reference architecture. Self-describes the harness's MCP wiring, witness signing, audit log, identity posture (always 'none' at v0.1). --check verifies an existing manifest, --dry-run prints without writing, --json emits to stdout.

  • 675 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added August 31, 2026
ai-agentsgobashsecurity

Works with

  • cli
  • mcp

Security analysis

A100/100

Scanned August 31, 2026

npx -y skills add ruvnet/metaharness --skill oia-manifest --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Oia Manifest?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Oia Manifest
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ruvnet-oia-manifest/badge)](https://www.skillsdirectory.com/skills/ruvnet-oia-manifest)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: oia-manifest
description: Emit .harness/oia-manifest.json declaring layer alignment with the OIA v0.1 9-layer reference architecture. Self-describes the harness's MCP wiring, witness signing, audit log, identity posture (always 'none' at v0.1). --check verifies an existing manifest, --dry-run prints without writing, --json emits to stdout.
---

# oia-manifest

> Codex skill: emit/validate the OIA cross-cutting manifest layer
> (ADR-034, iter 121 → iter 122).

## What it does

Emits `.harness/oia-manifest.json` — the harness's self-assessment against
the **Open Intelligence Architecture (OIA) v0.1** 9-layer reference model
published by the Agentics Foundation at <https://oia.agentics.org>.

The manifest declares:

- **9 layer alignments** — L1 physicalCompute → L9 humanAndBrowserInterface,
  each `full` / `partial` / `none` / `not-applicable`
- **6 horizontal spans** — security · observability · identity · governance ·
  policyEnforcement · interoperability, each `full` / `partial` / `none`
  with an `implementation` reference (file path or ADR number)
- **4 adjacent standards** — `mcp`, `a2a`, `acp`, `agentProtocol` with
  current mode + wiring notes
- **`discoveryEndpoint` + `registryUrl`** (both `null` at v0.1; OIA has no
  registry yet)

## Why a manifest, not a host adapter?

OIA at v0.1 has no runtime and no wire protocol — it's a vendor-neutral
vocabulary for assessing system alignment. The right plug-in shape is a
**static cross-cutting manifest** (ADR-034 Decision: Option B). It plugs in
ABOVE the host adapter contract and BELOW the MCP policy gate.

## Usage from Codex

```
/oia-manifest path=./my-harness
/oia-manifest path=./my-harness check=true
/oia-manifest path=./my-harness dry-run=true
```

## Equivalent CLI

```bash
harness oia-manifest ./my-harness                # write .harness/oia-manifest.json
harness oia-manifest ./my-harness --check        # validate existing manifest
harness oia-manifest ./my-harness --dry-run      # stdout, no file write
harness oia-manifest ./my-harness --json         # stdout JSON
```

## Verdict + exit codes (--check)

| Verdict | Exit | Meaning |
|---|---|---|
| `PASS` | 0 | manifest shape ok, oiaVersion echoed |
| `DRIFT` | 1 | shape mismatch — per-field reasons surfaced |
| `FAIL no manifest` | 2 | `.harness/oia-manifest.json` missing |

## Sample manifest (excerpt)

```json
{
  "schema": 1,
  "oiaVersion": "0.1",
  "harnessId": "my-bot@0.1.0",
  "layerAlignment": {
    "L4_toolsAndIntegrations": "full",
    "L7_governanceAndPolicy": "full"
  },
  "horizontalSpans": {
    "identity": { "status": "none", "implementation": null },
    "security": { "status": "full", "implementation": "mcp-policy.json + ADR-022" }
  },
  "adjacentStandards": {
    "mcp": { "mode": "local", "policyPath": ".harness/mcp-policy.json" }
  },
  "discoveryEndpoint": null,
  "registryUrl": null
}
```

## Pre-emptive composition rule (ADR-034 §120)

If a future OIA identity primitive ever wants to widen an MCP permission,
the composition is **denied at the policy gate**. The `mcp-policy.json`
default-deny posture is not negotiable for external identity claims.

## Related skills

- `validate-harness` (iter 22) — release-readiness umbrella
- `threat-model` (iter 114) — MCP threat-model for PR/compliance review
- `diag-harness` (iter 70) — kernel-version skew check

## See also

- [ADR-034 — OIA Integration](../../../docs/adrs/ADR-034-oia-integration.md)
- [ADR-022 — MCP primitive](../../../docs/adrs/ADR-022-mcp-primitive.md)
- [ADR-030 — Discovery Loop](../../../docs/adrs/ADR-030-discovery-loop.md)
- [oia.agentics.org](https://oia.agentics.org)

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.