QM: Scope-Based Agent Governance

Overview

This series explains how QM uses scope to coordinate identity, conversation context, state, credentials, compute, collaboration, automation, and audit. It follows a request from audience-aware resolution through scoped computers, memory, sessions, security gates, sharing, scheduled work, and deployment. The result is a practical way to decide what is writable, what crosses a boundary, what survives, and who receives an effect, while keeping model disclosure, provider behavior, and operator configuration as residual limits rather than treating architecture as a proof of safety.

Learning outcomes

QM makes scope the unit that joins identity, context, state, action, audience, and accountability.
A turn-local view can carry selected resources read-only while keeping the conversation's workspace writable and audience-bounded.
Memory, files, sessions, tape, sandboxes, and published artifacts provide different kinds of continuity; none should carry every durability promise.
Security posture, sharing, automation, and deployment controls narrow risk but do not make model output, providers, or operators fully trustworthy.

Keywords

QMScopeGovernanceAuthorizationSecurityCollaborationSandboxMemoryAutomationDeployment

Chapters

QM: Scope as Agency

Separate ownership from shared work

9 min · Separate ownership from shared work

9 min

QM: Turn Context Resolution

Separate writable work from visible context

10 min · Separate writable work from visible context

10 min

QM: Computer Lifecycle and Continuity

Separate files, memory, sessions, and tape

13 min · Separate files, memory, sessions, and tape

13 min

QM: Scope-to-State Model

Classify ownership, visibility, and persistence

9 min · Classify ownership, visibility, and persistence

9 min

QM: Layered Security Controls

Match each gate to its failure

12 min · Match each gate to its failure

12 min

QM: Controlled Sharing

Keep transfers read-only and audience-bounded

12 min · Keep transfers read-only and audience-bounded

12 min

QM: Automation and Deployment

Carry authority through time without hiding responsibility

12 min · Carry authority through time without hiding responsibility

12 min

QM: Capabilities and Limits

Read guarantees at the strength the evidence supports

15 min · Read guarantees at the strength the evidence supports

15 min

Preview every chapter · Unlock full text, audio, and video from chapter 4

Plans

Ch.1-3 FreeCh.4-8 Premium

FAQ

What problem is QM designed to solve?

QM is designed to let people share useful agent work without turning every personal context, credential, process, or hidden assumption into shared authority. Its scope model connects identity, audience, resources, actions, and accountability.

How does QM keep a shared conversation from inheriting private context?

QM resolves the full audience, admits only authorized resources, and carries cross-scope material as read-only, source-labelled handles. The model may still disclose a fact it legitimately saw, so authorization is not a perfect output filter.

How does QM keep agent work attributable over time?

QM ties work to an owner scope, records session and automation activity, and rechecks audience and authorization as work continues. Different stores preserve different state, while sandbox recovery and provider behavior remain bounded.

Is QM a complete security guarantee for agent work?

No. QM supplies scope resolution, capability and audience checks, screening, approval, egress paths, and audit, but model disclosure, heuristic screening, plaintext credentials in processes, provider behavior, bearer links, and operator configuration remain limits.