Mx-Modes: A Meta-Harness Framework for Multi-Mode AI Operation

Mx-Modes is a meta-harness framework that structures AI operation through explicit mode orientation. A root mode.md file activates a governing harness composed of five discipline-bearing surfaces — Mind, Morals, Mission, Memory, and Methods — plus the execution surface Means. These surfaces separate reasoning discipline, permission boundaries, operational purpose, contextual continuity, codified tradecraft, and execution capability. The framework treats the model as a probabilistic generator and the harness as the governing architecture: structured separation of concerns at the operating-envelope layer, not at the prompt or fine-tune layer.

The construct is harness-agnostic and scale-invariant — it composes from per-agent deployments up through multi-agent and enterprise-scale operations — and bracketed at the attach end by the root mode file (the harness-specific entry, swappable) and at the execution end by Means (the substrate-specific execution surface, swappable). The five discipline-bearing surfaces are the durable, harness-agnostic core that survives both swaps.

The Methods surface holds the codified, evidenced, reusable procedures and rigor disciplines for doing work to a standard. Its boundary with Morals is enforcement-and-obligation: Morals holds the obligatory, enforced disciplines (violation is a breach); Methods holds recommended craft (departure is suboptimal, not a violation). The two compose through a graduation pipeline: a repeated practice is codified as a method; a method that proves it must be mandatory graduates to a moral; an obligation that can be mechanized graduates to means (an enforcement hook). This makes "enforcement over documentation" a lifecycle rather than a slogan.

This v2 release is a substantive expansion of the v1 deposit. v2 (2026-05-27): Adds the Methods discipline surface (§3.5) per ADR-EA-0026. Surface counts updated throughout (five → six; four discipline-bearing → five). Section 8 (Architectural Value) extended; abstract revised. Ratified by founder call (JD Longmire, 2026-05-27) and integrated across the AIDE canon (aide-canon PR #42 + PR #43), the NG-AIDE-01 reference instance (PR #51), and the thinx-Claude operator-altitude reference deployment.

The framework remains co-authored with Micah Longmire per ADR-EA-0008 corpus-level identity. Canonical source: github.com/ologos-repos/aide-canon/tree/main/constructs/mxm.