Future Directions
Governance v4 — Future Directions
Governance v4 is not implemented. The items below are future architectural directions identified during Governance v3 development. They are recorded to preserve intent, not as implementation commitments.
Behavioral Authority Chain
Current Governance inventories behavioral sources. A future Governance v4 may expose the complete behavioral authority chain used by an agent, including provider system prompts, developer prompts, workspace harnesses, runtime prompt injection, prompt templates, dynamically injected context, operator instructions, precedence between instruction layers, and active vs discovered behavioral sources.
Long-term objective: transform expert knowledge about AI harnesses into observable operational information, allowing operators to understand why an agent behaves as it does without requiring deep knowledge of provider-specific prompt architectures.
Current Governance inventories behavioral sources. A future Governance v4 may expose the complete behavioral authority chain used by an agent, including provider system prompts, developer prompts, workspace harnesses, runtime prompt injection, prompt templates, dynamically injected context, operator instructions, precedence between instruction layers, and active vs discovered behavioral sources.
Long-term objective: transform expert knowledge about AI harnesses into observable operational information, allowing operators to understand why an agent behaves as it does without requiring deep knowledge of provider-specific prompt architectures.
Start Here
Governance v3 inspects knowledge sources and behavioral sources.
This page inspects existing operational traces and active harness candidates: docs, ADRs, implementation records, checkpoints, logs, workspaces, orientation narratives, backup traces, AGENTS.md files, CLAUDE.md files, prompt templates, and other discoverable instruction layers.
Documentation governs intent. Behavioral sources govern execution. Governance v3 makes both visible so Instruction Drift can be detected before active harnesses diverge from canonical decisions.
Waiting for evidence...