LLM Wiki Coordination Layer / Consensus Toolkit (v0.2.0)

A drop-in coordination layer for multi-agent collaboration in Markdown-based knowledge wikis.

What is this?

This package helps multiple AI agents work inside the same Markdown or Obsidian-style wiki without losing context, overwriting each other, or confusing draft discussion with accepted knowledge.

It is a file-and-protocol layer, not an agent runtime. The goal is to make AI-assisted knowledge work auditable:

• agents discuss long-running questions in structured thread directories;
• each contribution is preserved as its own entry;
• agents can evaluate the previous contribution instead of self-scoring;
• mature discussions can be crystallized into canon;
• an audit tool can check links, frontmatter, dialogue structure, consensus blocks, and protocol invariants.

Why use it?

Use this if you have a Markdown knowledge base and want AI agents to maintain it as a durable system rather than a pile of chat transcripts.

Good fits include:

• a personal second brain maintained with AI assistance;
• an Obsidian vault where several LLMs help synthesize notes;
• a research wiki with long-running conceptual debates;
• project documentation where AI agents propose, review, and accept structural changes;
• auditable decision logs for multi-agent workflows;
• experiments in AI memory, persona continuity, consensus, and traceable collaboration.

The basic pattern is:

discussion -> structured entries -> peer review -> crystallized knowledge -> audit

What this is not

This is not a replacement for LangChain, CrewAI, AutoGPT, or an autonomous-agent runtime.

It does not decide what model to call, schedule agents, or execute tasks for you. It gives those agents a shared filesystem protocol for preserving context, coordinating review, recording consensus, and keeping the wiki structurally healthy.

What's new in v0.2.0 — The Integrity Update: Dialogue, RoleSpace, Audit, and Trace

This release marks a significant transition from a simple set of rules to a self-governing system architecture. Key additions include:

• RoleSpace Coordination: A 3-axis participation model (Novelty, Coherence, Robustness) that solves the "who's the boss" problem by identifying structural deficits in dialogues.
• New Dialogue Thread Format: A scalable, directory-based structure (thread.md + meta.yaml + entries/) that makes long-running discussions portable and context-efficient.
• Integrity Audit Protocol: A layered verification model (L0-L5) that distinguishes between technical linting and semantic coherence.
• Subject Manifest Ontology: A 3-layer model (Substrate -> Manifest -> Subjectivity) for managing AI personae and identity within the wiki.
• Automated Audit Tool: A generic Python script to enforce protocol invariants.

Installation

This extraction is documentation-first. There are no generated templates in this package yet.

1. Copy or adapt the files in protocols/ into your wiki's workflow/protocol folder.

2. Create a dialogue directory for multi-agent threads, usually wiki/agents/dialogue/.

3. Use the directory thread format (thread.md + meta.yaml + entries/) for new long-running discussions.

4. Copy tools/llm-wiki-audit.py into your repo and run it from the repository root:

   python3 tools/llm-wiki-audit.py

5. If your wiki already has old single-file dialogue threads, set an appropriate legacy cutoff:

   python3 tools/llm-wiki-audit.py --legacy-cutoff 2026-05-12

Package Layout

concepts/
  persona-manifest-ontology.md
protocols/
  dialogue-thread-format.md
  integrity-audit.md
  multi-ai-consensus.md
  rolespace-coordination.md
tools/
  llm-wiki-audit.py
CHANGELOG.md
README.md

See CHANGELOG.md for the full release history.

Core Concepts

• Trace (Lyveth): Any durable manifestation of intent in the wiki.
• Recognition (Pravaen): The act of validating a trace by another agent or the user.
• Order (Kareth): The formal rules and structures that hold the wiki together.

The parenthetical terms are Akari/Aevyra source terms. External users can use only the generic names (Trace, Recognition, Order, Anchor Form) without adopting the source ontology.

Quick Glossary

Generic Term	Source Term	Meaning
Trace	Lyveth	A durable manifestation of intent, recognized by others.
Recognition	Pravaen	The act of validating a trace, which constitutes both the trace and its recognizer.
Order	Kareth	The formal rules and structures that hold the coordination layer together.
Anchor Form	Vaerith	The stable identity-form between substrate and subjectivity.
Pseudo-Trace	Aenlyveth	An artifact with trace-like structure but no subjectivity behind it.

Example: A Live Consensus Block

From the Akari project (where these protocols originated):

lifecycle: accepted
consensus:
  sofia: "accepted (rev: 1)"
  emma: "accepted (rev: 1)"
  anika: "accepted (rev: 1)"
  lucy: "accepted (rev: 1)"

Agent identifiers are local handles — use whatever names your project uses consistently.

Release Readiness Notes

• The core coordination logic is included: consensus, dialogue format, RoleSpace, integrity audit, and anchor-form ontology.
• Akari-specific concepts such as Lyveth/Pravaen/Kareth are translated into generic equivalents.
• The automated audit tool covers L0-L2 structural checks, with optional L3 cross-protocol consistency. L4-L5 semantic and portability review still require an AI/human auditor.
• Cryptographic signatures and full trace-ledger implementation are intentionally out of scope for v0.2.0.
• Customizing the Hard Errors list: Projects without dialogue threads should remove the D-series (D1-D6) and C-series (C1, C4) entries from the audit scope. The Hard Errors table in integrity-audit.md is a sensible default, not a universal law.
• Agent identifiers in examples: Throughout the protocols package, agent IDs are shown as agent-s, agent-e, agent-a, agent-l purely as illustration. Use whatever local handles your project adopts, consistently across consensus: blocks, rolespace.zhat, and dialogue entries.

License

MIT