Proposal: Workflow-aware context layer on top of OKF for multi-agent coordination

[JTM0115]

First of all, thank you for publishing the Open Knowledge Format specification. I find the direction very relevant to a problem I have been exploring in AI agent-based software development and operations.

This is not a commercial proposal, but a non-confidential technical idea that may be useful for the broader discussion around agent-readable knowledge, context portability, and workflow interoperability.

My understanding is that OKF focuses on a simple and portable way to represent knowledge as Markdown files with structured metadata. This is valuable because agents need context that is both human-readable and machine-consumable.

However, in practical multi-agent development workflows, agents often need more than static knowledge documents. They also need a way to understand project state, task ownership, status transitions, trust boundaries, and cross-project coordination rules.

In my own work, I have been referring to this broader coordination layer as an Agent Context Hub: a lightweight layer that connects project knowledge, task state, workflow transitions, trust boundaries, and agent-facing context into a more operational structure.

A possible extension direction could include:

1. Agent-readable context documents
   Project status, architecture notes, API contracts, decisions, risks, and operational constraints could be represented as structured Markdown concepts compatible with OKF-style metadata.

2. Source-file-level context headers
   Important source files could include compact agent-facing headers describing the file role, responsibilities, invariants, reusable utilities, and prohibited modifications. This helps coding agents avoid duplicate implementation, unnecessary file fragmentation, and context drift.

3. Workflow-aware context objects
   In addition to static documents, agents often need structured objects such as tasks, board posts, review requests, implementation notes, and decision records. These objects benefit from explicit states, versioning, optimistic locking, and idempotency.

4. Trust and governance boundaries
   Agent systems need to distinguish trusted project instructions from untrusted external content. A practical standard may need concepts such as trusted context, untrusted evidence, policy-guarded actions, role-based permissions, and auditable state transitions.

5. Cross-project coordination
   In real development environments, one project often depends on another. A portable context model could support cross-project references, dependency requests, review handoffs, and machine-readable task/status exchange between agents or teams.

The key distinction is that OKF appears to address how knowledge is represented and exchanged, while this proposed layer focuses on how agents coordinate work using that knowledge.

I see these as complementary layers rather than competing approaches:

• OKF-compatible Markdown bundles for portable knowledge exchange
• Additional metadata for project state, task state, decision records, and trust classification
• A lightweight API or protocol layer for state transitions, review workflows, and audit logs
• Agent-facing rules that prevent direct modification of protected contracts or project invariants

I would be interested to hear whether this kind of workflow-aware context layer is within the intended scope of future OKF extensions, or whether it would be better treated as a separate companion specification built on top of OKF.