Starter Bundle

---
type: "Reference"
title: "LLM Wiki README"
description: "Act as the front door for humans and AI agents by defining the wiki scope, reading order, ownership, and safe starting workflow."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Knowledge steward"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly or when the wiki scope changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Knowledge steward"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly or when the wiki scope changes"
audience: "humans and AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# LLM Wiki README

## Purpose
Act as the front door for humans and AI agents by defining the wiki scope, reading order, ownership, and safe starting workflow.

## What To Write
- Name the domain this LLM Wiki covers and the domains it does not cover.
- Link the first five pages a new maintainer should read.
- State where authoritative decisions, policies, runbooks, and open questions live.
- Include a one-paragraph quick start for humans and a one-paragraph quick start for AI agents.

Example:

This wiki covers the Acme Billing Platform: product context, system architecture, release operations, support procedures, decisions, policies, and approved agent instructions. It does not contain secrets, customer records, raw incident chat logs, legal advice, or unreviewed strategy drafts.

## Owner And Review Cadence
- Owner role: Knowledge steward
- Review frequency: monthly or when the wiki scope changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Read this first. Use it to decide whether the wiki is in scope for the current task, then follow the linked index, governance, and agent rules before making edits.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "LLM Wiki Index"
description: "Provide deterministic navigation so humans and agents can route through the wiki without scanning every file."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Documentation lead"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "every approved content change"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Documentation lead"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "every approved content change"
audience: "humans, AI agents, search systems"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# LLM Wiki Index

## Purpose
Provide deterministic navigation so humans and agents can route through the wiki without scanning every file.

## What To Write
Create sections for organization, products, architecture, operations, decisions, policies, agent rules, and onboarding. Each entry should include path, summary, owner, status, sensitivity, last reviewed date, source identity, discovery surface, and related pages. For durable reports, coding standards, architecture, best practices, and preserved handoff records, the summary is routing-only: one paragraph that explains why to open the page without copying the body.

For one project, this INDEX can be the local routing index. For a multisite or multi-project LLM Wiki, keep the root INDEX to project namespaces, links to each project-level INDEX and LOG counterpart, and global-only files such as coding standards, organization policy, governance, source maps, and workspace.uai. Put each project's scoped routing detail inside its own sub-wiki index and each project's activity history inside its own sub-wiki log.

Example row:

| Path | Routing summary | Owner | Status | Sensitivity | Last reviewed | Source identity | Discovery surface | Related |
|---|---|---|---|---|---|---|---|---|
| architecture/SYSTEM_OVERVIEW.md | Routes agents to current system boundaries and service ownership before architecture or runbook changes. | Platform Team | authoritative | internal | 2026-04-28 | sha256:... | llm-wiki/INDEX.md | architecture/SERVICES.md, decisions/DECISION_LOG.md |

## Owner And Review Cadence
- Owner role: Documentation lead
- Review frequency: every approved content change
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this as the routing table. Select the smallest useful page set before opening deeper files, and update it whenever durable pages are added, renamed, archived, or reclassified.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "LLM Wiki Governance"
description: "Define ownership, review, approval, archival, and escalation rules for the wiki."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Knowledge governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "quarterly and after governance changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Knowledge governance owner"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "quarterly and after governance changes"
audience: "maintainers, reviewers, contributors, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# LLM Wiki Governance

## Purpose
Define ownership, review, approval, archival, and escalation rules for the wiki.

## What To Write
Document page ownership, required reviewers, approval thresholds, review cycles, archival rules, and what counts as a high-impact change. Name who can approve policy, security, privacy, architecture, production, and public-facing changes.

Recommended sections:
- Ownership model
- Review cycles by page type
- Approval matrix
- AI-generated edit policy
- Deprecation and archive process
- Escalation path

## Owner And Review Cadence
- Owner role: Knowledge governance owner
- Review frequency: quarterly and after governance changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this to decide whether an edit can be made directly, proposed as a draft, or blocked until a human approves it.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "LLM Wiki Trust Model"
description: "Explain the trust labels, sensitivity labels, source-of-truth rules, and agent permissions used across the wiki."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Governance and security reviewers"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "quarterly or when labels change"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Governance and security reviewers"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "quarterly or when labels change"
audience: "humans, AI agents, auditors"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# LLM Wiki Trust Model

## Purpose
Explain the trust labels, sensitivity labels, source-of-truth rules, and agent permissions used across the wiki.

## What To Write
Define these labels and how agents should interpret them:
- authoritative: current source of truth inside this wiki.
- working-draft: useful but not approved.
- proposal: candidate future state.
- needs-review: stale, incomplete, conflicting, or ownerless.
- historical: context only.
- deprecated: superseded and unsafe to follow for current work.
- external-reference: useful outside source, not owned by this wiki.

Also define sensitivity labels such as public, internal, confidential, restricted, and never-store.

## Owner And Review Cadence
- Owner role: Governance and security reviewers
- Review frequency: quarterly or when labels change
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use labels exactly. Never treat a draft, proposal, historical note, deprecated page, or external reference as approved policy.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "LLM Wiki Contributing Guide"
description: "Tell humans and AI agents how to propose useful changes without turning the wiki into an unreviewed dump."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Documentation lead"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "quarterly or when workflow changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Documentation lead"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "quarterly or when workflow changes"
audience: "contributors, maintainers, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# LLM Wiki Contributing Guide

## Purpose
Tell humans and AI agents how to propose useful changes without turning the wiki into an unreviewed dump.

## What To Write
Describe the contribution path:
1. Read README, INDEX, GOVERNANCE, TRUST_MODEL, and relevant page owners.
2. Classify the change as correction, update, new page, archive, policy change, or sensitive change.
3. Preserve citations and provenance.
4. Add or update related links.
5. Request the correct reviewer.

State what is not accepted: secrets, customer records, unapproved policy rewrites, uncited claims, and vague bulk rewrites.

## Owner And Review Cadence
- Owner role: Documentation lead
- Review frequency: quarterly or when workflow changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this before changing files. Prefer narrow edits, preserve citations, add review notes, and avoid mixing unrelated fixes.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "LLM Wiki Changelog"
description: "Record durable wiki changes so future maintainers can understand what changed, why, and who reviewed it."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Knowledge steward"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "every release or approved change batch"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Knowledge steward"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "every release or approved change batch"
audience: "maintainers, auditors, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# LLM Wiki Changelog

## Purpose
Record durable wiki changes so future maintainers can understand what changed, why, and who reviewed it.

## What To Write
Use dated entries with changed pages, reason, reviewer, source references, and follow-up tasks.

Example:

## 2026-04-28
- Updated architecture/SYSTEM_OVERVIEW.md to reflect the new billing-worker boundary.
- Added decision DEC-004 on asynchronous invoice retries.
- Reviewer: Platform Team lead.
- Follow-up: update operations/RUNBOOKS.md before the next release.

## Owner And Review Cadence
- Owner role: Knowledge steward
- Review frequency: every release or approved change batch
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Append entries. Do not rewrite history except to correct factual mistakes with a note.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "LLM Wiki Glossary"
description: "Define organizational terms, acronyms, product names, service names, roles, and domain concepts in a way agents can cite."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Domain owner or documentation lead"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and when terminology changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Domain owner or documentation lead"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and when terminology changes"
audience: "new humans, AI agents, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# LLM Wiki Glossary

## Purpose
Define organizational terms, acronyms, product names, service names, roles, and domain concepts in a way agents can cite.

## What To Write
Each entry should include term, plain definition, owner, related pages, and forbidden interpretations.

Example:

## Ledger Event
Authoritative definition: An immutable event emitted by the billing ledger after validation.
Owner: Billing Platform Team.
Do not infer: A ledger event is not the same as a payment gateway webhook.
Related: architecture/DATA_FLOWS.md, operations/RUNBOOKS.md

## Owner And Review Cadence
- Owner role: Domain owner or documentation lead
- Review frequency: monthly and when terminology changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use glossary definitions when interpreting ambiguous names. Flag unknown or conflicting terms instead of guessing.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "System Overview"
description: "Describe the current system boundary, main components, data flows, dependencies, ownership, and known gaps."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Platform or architecture owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "quarterly and after major architecture changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Platform or architecture owner"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "quarterly and after major architecture changes"
audience: "engineers, operators, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# System Overview

## Purpose
Describe the current system boundary, main components, data flows, dependencies, ownership, and known gaps.

## What To Write
Include:
- Current system purpose and non-goals.
- Component map with owners.
- Critical data flows.
- External dependencies.
- Deployment environments.
- Current constraints and known gaps.
- Links to architecture decisions and runbooks.

Do not bury current state and history together. Use separate sections for current architecture, historical context, and planned changes.

## Owner And Review Cadence
- Owner role: Platform or architecture owner
- Review frequency: quarterly and after major architecture changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this for high-level orientation, then open service, data-flow, dependency, and decision pages before making implementation claims.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Architecture Decisions"
description: "Keep durable architecture decisions close to the system pages they affect."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Architecture owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "when decisions are proposed, approved, superseded, or retired"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Architecture owner"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "when decisions are proposed, approved, superseded, or retired"
audience: "architects, engineers, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Architecture Decisions

## Purpose
Keep durable architecture decisions close to the system pages they affect.

## What To Write
Use one entry per decision:
- ID and title
- Status: proposed, accepted, superseded, deprecated
- Date
- Owner and approvers
- Context
- Decision
- Alternatives considered
- Consequences
- Related pages

Agents may draft new decision records, but acceptance requires human approval.

## Owner And Review Cadence
- Owner role: Architecture owner
- Review frequency: when decisions are proposed, approved, superseded, or retired
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this to avoid re-litigating settled tradeoffs. Preserve rejected options and supersession notes.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Decision Log"
description: "Record product, policy, operational, and architecture decisions that should survive chat history, tickets, and personnel changes."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Decision owner or accountable lead"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "whenever a durable decision is made"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Decision owner or accountable lead"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "whenever a durable decision is made"
audience: "humans, AI agents, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Decision Log

## Purpose
Record product, policy, operational, and architecture decisions that should survive chat history, tickets, and personnel changes.

## What To Write
Each decision should include:
- Decision ID
- Date
- Status
- Decision owner
- Problem
- Decision
- Rationale
- Rejected options
- Follow-up work
- Related pages

Use explicit statuses: proposed, accepted, superseded, rejected, blocked, reopened.

## Owner And Review Cadence
- Owner role: Decision owner or accountable lead
- Review frequency: whenever a durable decision is made
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use approved decisions as constraints. Do not collapse open disagreement into a fake consensus.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Open Questions"
description: "Keep uncertainty visible instead of letting agents invent answers or treat gaps as settled facts."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Relevant domain owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "weekly during active work; monthly otherwise"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Relevant domain owner"
status: "Proposal/Draft"
trust_level: "working-draft"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "weekly during active work; monthly otherwise"
audience: "owners, reviewers, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: false
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Open Questions

## Purpose
Keep uncertainty visible instead of letting agents invent answers or treat gaps as settled facts.

## What To Write
Track:
- Question
- Why it matters
- Current owner
- Needed evidence
- Blocked pages or decisions
- Target review date
- Resolution when closed

Agents may add questions when sources conflict, owners are missing, or a requested action lacks approval.

## Owner And Review Cadence
- Owner role: Relevant domain owner
- Review frequency: weekly during active work; monthly otherwise
- Human review required for updates: false
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Check this before answering uncertain questions. If a page depends on an unresolved issue, cite the open question.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Operations Runbooks"
description: "Provide safe, current operating procedures with triggers, steps, checks, rollback paths, and escalation rules."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Operations owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after incidents or process changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Operations owner"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after incidents or process changes"
audience: "operators, support staff, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Operations Runbooks

## Purpose
Provide safe, current operating procedures with triggers, steps, checks, rollback paths, and escalation rules.

## What To Write
Each runbook should include:
- Trigger
- Severity or scope
- Preconditions
- Step-by-step procedure
- Verification checks
- Rollback or stop condition
- Required approvals
- Escalation path
- Last successful use

Make dangerous actions explicit and state which actions require a human.

## Owner And Review Cadence
- Owner role: Operations owner
- Review frequency: monthly and after incidents or process changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Agents may summarize or check runbooks. They must not execute production actions, approve releases, or bypass human authorization.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agent Instructions"
description: "Define how AI agents should read, cite, update, and respect the LLM Wiki."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after agent workflow changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance owner"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after agent workflow changes"
audience: "AI agents and maintainers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agent Instructions

## Purpose
Define how AI agents should read, cite, update, and respect the LLM Wiki.

## What To Write
State:
- Required first-read pages.
- Which paths are read-only.
- Which paths are writable by agents.
- How to identify authoritative pages.
- How to handle stale, conflicting, deprecated, or draft content.
- Final response requirements.
- Actions requiring explicit human approval.

Agents must preserve citations, flag uncertainty, avoid secrets, and never treat drafts as approved policy.

## Owner And Review Cadence
- Owner role: AI governance owner
- Review frequency: monthly and after agent workflow changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
This is the operational contract. Read it before editing. Follow the retrieval, update, citation, and safety boundary files it links.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agent Retrieval Guide"
description: "Tell agents how to choose the smallest useful page set and how to route through the wiki before answering."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly or when navigation changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance owner"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly or when navigation changes"
audience: "AI agents and retrieval-system builders"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agent Retrieval Guide

## Purpose
Tell agents how to choose the smallest useful page set and how to route through the wiki before answering.

## What To Write
Define the retrieval order:
1. README
2. INDEX
3. Trust model
4. Relevant authoritative pages
5. Related decisions and policies
6. Raw sources only when needed

Describe how to handle contradictions, missing pages, draft pages, deprecated pages, and external references.

## Owner And Review Cadence
- Owner role: AI governance owner
- Review frequency: monthly or when navigation changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Read INDEX first, prefer authoritative pages, include citations, and open raw sources only when compiled pages are insufficient or need verification.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agentic Orchestration Runbook"
description: "Define how agentic agents and orchestration layers should use the LLM Wiki before, during, and after a run."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after orchestration workflow changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance owner"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after orchestration workflow changes"
audience: "AI agents, orchestrator operators, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agentic Orchestration Runbook

## Purpose
Define how agentic agents and orchestration layers should use the LLM Wiki before, during, and after a run.

## What To Write
Before an orchestrated run:
- Name the task boundary: answer, draft, code change, audit, release, or wiki update.
- Read README, INDEX, TRUST_MODEL, AGENT_INSTRUCTIONS, RETRIEVAL_GUIDE, and SAFETY_BOUNDARIES.
- Build a compact task packet with facts, constraints, owners, source links, forbidden actions, and review gates.
- Declare whether agents are read-only, proposal-only, staged-write, or human-approved write.
- Name where run notes, tool outputs, traces, evaluations, proposed changes, and open questions should be saved.

During the run:
- Keep working notes separate from reviewed wiki pages.
- Cite wiki pages, raw sources, and canonical external sources for important claims.
- Preserve stale claims, contradictions, missing owners, and uncertainty as task state.
- Treat runtime traces, tool outputs, and agent messages as evidence inputs until reviewed.

After the run:
- Stage reusable answers or syntheses as wiki drafts with source links and review status.
- Save trace, evaluation, or tool outputs only in approved evidence locations.
- Promote facts into reviewed wiki pages only after source, privacy, owner, contradiction, and freshness checks.
- Keep public MCP, A2A, write API, trace exporter, live evaluation, adapter, SDK, CLI, certification, and managed-service language marked as planned unless current public evidence exists.

## Owner And Review Cadence
- Owner role: AI governance owner
- Review frequency: monthly and after orchestration workflow changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this before orchestrated runs. Build a compact task packet, preserve trace and tool outputs as evidence inputs, stage proposed wiki updates, and stop at review boundaries.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agentic Task Packet Template"
description: "Provide a compact, reviewable packet for one agentic run without loading the whole wiki or hiding review boundaries."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after orchestration workflow changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance owner"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after orchestration workflow changes"
audience: "orchestrator operators, AI agents, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agentic Task Packet Template

## Purpose
Provide a compact, reviewable packet for one agentic run without loading the whole wiki or hiding review boundaries.

## What To Write
Use one task packet per orchestrated run.

## Task Boundary

- Run name:
- Requested outcome:
- Task type: answer, draft, code change, audit, release support, wiki update, other
- In scope:
- Out of scope:
- Stop conditions:

## Required Reading

- Wiki pages to read:
- Raw/source records to inspect:
- External canonical sources:
- Pages explicitly not needed:

## Permissions

- Agent mode: read-only, proposal-only, staged-write, human-approved write
- Writable paths:
- Read-only paths:
- Human approval required for:
- Forbidden actions:

## Context Packet

- Current facts:
- Constraints:
- Owners:
- Sensitive-data boundaries:
- Known contradictions:
- Stale or source-needed claims:

## Evidence Destinations

- Run notes path:
- Proposed wiki updates path:
- Tool outputs path:
- Trace or evaluation path:
- Review queue path:
- Roadmap or open-question path:

## Completion Report

- Pages read:
- Sources cited:
- Artifacts changed or proposed:
- Evidence produced:
- Checks run:
- Review questions:
- Follow-up tasks:

Do not treat this packet as durable truth by itself. Promote useful facts only after source, owner, privacy, contradiction, freshness, and verification checks.

## Owner And Review Cadence
- Owner role: AI governance owner
- Review frequency: monthly and after orchestration workflow changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Fill this before an orchestrated run. Keep it small, cite wiki/source paths, name permissions, and record where evidence and proposed updates should land.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agentic Support Escalation Checklist"
description: "Give agents and reviewers a bounded support path when an orchestrated run hits conflicts, missing authority, privacy limits, stale sources, or tool failures."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after support workflow changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance owner"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after support workflow changes"
audience: "AI agents, orchestrator operators, support reviewers, maintainers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agentic Support Escalation Checklist

## Purpose
Give agents and reviewers a bounded support path when an orchestrated run hits conflicts, missing authority, privacy limits, stale sources, or tool failures.

## What To Write
Use this when an orchestrated run needs support instead of more autonomous work.

## Escalation Triggers

Escalate before continuing when the run hits:

- A conflict between reviewed wiki pages or canonical sources.
- A source-needed, stale, deprecated, or ownerless claim that affects the answer or edit.
- Secrets, private transcripts, customer data, regulated data, or unclear consent.
- Security, privacy, legal, production, financial, medical, or policy decisions.
- Tool failure, partial execution, missing logs, missing tests, or unverifiable output.
- A requested action outside the task packet or outside current agent permissions.
- A claim about public MCP, A2A, write API, trace exporter, live evaluation, adapter, SDK, CLI, certification, managed service, or official support without current public evidence.

## First Response

- Stop the unsafe or unsupported action.
- Preserve current notes, tool outputs, citations, and error messages in the approved evidence destination.
- Name the trigger in plain language.
- Separate known facts from assumptions, proposals, and open questions.
- Do not edit reviewed wiki pages until the escalation is resolved.

## Support Packet

- Run name:
- Task packet path:
- Trigger:
- Pages read:
- Sources cited:
- Tool outputs or traces:
- Sensitive-data concern:
- Contradiction or stale claim:
- Proposed next action:
- Human reviewer needed:
- Checks already run:
- Follow-up owner:

## Resolution Paths

- If sources conflict, stage both claims with citations and ask the owner which source wins.
- If authority is missing, create an open question instead of inventing policy.
- If private data appears, stop, minimize exposed detail, and ask the data owner or privacy reviewer.
- If a tool fails, record exact command, output, environment, and retry status before proposing another run.
- If the request widens support claims, keep the language planned, candidate, or blocked until implementation evidence exists.
- If the task is complete but reusable, stage a reviewed wiki update with source links and checks.

## Completion Note

End the escalation with:

- Resolution:
- Artifacts updated:
- Evidence preserved:
- Wiki pages proposed:
- Human approvals:
- Remaining blockers:

This checklist is support guidance for governed wiki operations. It does not create a public runtime, support service, MCP server, A2A implementation, write API, trace exporter, live evaluation integration, SDK, CLI, certification, endorsement, or managed-service claim.

## Owner And Review Cadence
- Owner role: AI governance owner
- Review frequency: monthly and after support workflow changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this when a run cannot safely continue. Preserve evidence, classify the issue, stop unsafe work, and ask for the right human review instead of widening support claims.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Long-Run Agent Wiki Strategy"
description: "Define how extended agent runs use folded state, active subgoals, review-gated synthesis, and bounded wiki retrieval."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after long-run workflow changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance owner"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after long-run workflow changes"
audience: "AI agents, reviewers, maintainers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Long-Run Agent Wiki Strategy

## Purpose
Define how extended agent runs use folded state, active subgoals, review-gated synthesis, and bounded wiki retrieval.

## What To Write
Use this profile when an agent run is too long for one hot context window.

## Context-Folded Agent Run

- Macro Context: stable goal, scope, owner, authority boundary, support boundary, and stop conditions.
- Active Subgoal: the one current unit of work with done criteria.
- Micro Context: smallest useful page set, source records, files to touch, checks, blockers, and citations.
- Completed Subgoal Summary: compact record of finished work with changed artifacts, source traces, checks, blockers, and next dependencies.
- State Review: explicit decision to continue, fold, ask for review, or stop.

## Folded Run State Rule

Folded Run State records observable task state only. It must not contain private chain-of-thought, hidden reasoning, secrets, private raw data, or unreviewed authority claims.

## Active Wiki Navigation

1. Start from README, INDEX, source map, route manifest, log, or llms.txt.
2. Narrow to the smallest useful page set for the Active Subgoal.
3. Inspect owner, status, source_status, sensitivity, last_reviewed, and agent_use.
4. Cite local wiki pages and source records.
5. Stop when evidence, permission, privacy, or authority is unclear.

## Governed Agent Synthesis

Reusable synthesis begins as an Agent-Staged Page in `proposals/agent-staged/`. Promote only after source trace, owner review, lint, contradiction checks, and promotion notes exist.

## Owner And Review Cadence
- Owner role: AI governance owner
- Review frequency: monthly and after long-run workflow changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this before extended wiki work. Keep one Active Subgoal, write Folded Run State packets, cite local pages, and stage reusable synthesis for review.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Process-Quality Checks"
description: "Provide deterministic checks for long-running agent wiki work before summaries, staged pages, or promotion claims are accepted."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after process failures"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance owner"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after process failures"
audience: "AI agents, reviewers, maintainers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Process-Quality Checks

## Purpose
Provide deterministic checks for long-running agent wiki work before summaries, staged pages, or promotion claims are accepted.

## What To Write
Use these checks at each State Review.

## Fail Conditions

- Fabricated paths or tools: a route, file, script, command, package, upload, or verification is claimed without evidence.
- Unsupported success claims: publication, upload, route verification, certification, sync, hosted import, autonomous publication, or automation is claimed without proof.
- Repetitive loops: discovery repeats without a changed Active Subgoal, new evidence, or a blocker.
- Malformed packets: Folded Run State lacks Active Subgoal, citations, blockers, checks, or done criteria.
- Overbroad retrieval: the run scans broad archives, raw dumps, generated history, or sibling namespaces without explicit need.
- Missing source traces: Completed Subgoal Summary or Agent-Staged Page lacks local page/source paths.
- Private raw data exposure: sensitive source material appears in public pages, discovery files, or hot packets.
- Unreviewed synthesis as authority: Agent-Staged Pages or summaries are promoted without owner review and checks.

## State Review Report

- Active Subgoal:
- Evidence checked:
- Files changed or staged:
- Checks run:
- Blockers:
- Source traces:
- Completed Subgoal Summary path:
- Next action:

## Owner And Review Cadence
- Owner role: AI governance owner
- Review frequency: monthly and after process failures
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Run these checks at every State Review and before promoting Agent-Staged Pages.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agent Update Rules"
description: "Define when agents may edit, propose, stage, or refuse wiki updates."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Knowledge governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly or when permissions change"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Knowledge governance owner"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly or when permissions change"
audience: "AI agents, maintainers, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agent Update Rules

## Purpose
Define when agents may edit, propose, stage, or refuse wiki updates.

## What To Write
Allowed without approval:
- Fix broken internal links.
- Add missing related links.
- Add clearly labeled open questions.
- Draft source summaries from approved sources.

Requires human review:
- Policy changes.
- Security or privacy guidance.
- Architecture decisions.
- Production runbooks.
- Pages marked authoritative.
- Any edit involving sensitive data.

Forbidden:
- Adding secrets.
- Inferring permissions.
- Silently rewriting approved policy.
- Removing disagreement to make text cleaner.

## Owner And Review Cadence
- Owner role: Knowledge governance owner
- Review frequency: monthly or when permissions change
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Apply these rules before writing. If an update crosses a policy, security, privacy, legal, production, or source-of-truth boundary, ask for human approval.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agent Citation Rules"
description: "Make answers and updates traceable by requiring page citations, source paths, and uncertainty labels."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Documentation lead"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "quarterly or when citation standards change"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Documentation lead"
status: "Proposal/Draft"
trust_level: "reviewed"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "quarterly or when citation standards change"
audience: "AI agents, reviewers, auditors"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agent Citation Rules

## Purpose
Make answers and updates traceable by requiring page citations, source paths, and uncertainty labels.

## What To Write
Citation requirements:
- Cite the wiki page path for every important internal claim.
- Cite the source-summary or raw path for evidence-sensitive claims.
- Do not cite memory, vibes, or "common knowledge" for policy, architecture, security, privacy, legal, model, benchmark, or production statements.
- When sources disagree, cite both and label the conflict.

Answer format:
Claim. Source: `architecture/SYSTEM_OVERVIEW.md`; evidence: `raw/architecture/2026-04-platform-review.md`.

## Owner And Review Cadence
- Owner role: Documentation lead
- Review frequency: quarterly or when citation standards change
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Cite local wiki pages for internal knowledge, cite raw/source records for evidence, and state when a claim is unsupported, stale, or contradicted.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Agent Safety Boundaries"
description: "Protect sensitive data, permissions, production actions, and high-impact decisions from unsafe agent behavior."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Security and governance owners"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after incidents or policy changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Security and governance owners"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after incidents or policy changes"
audience: "AI agents, maintainers, security reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Agent Safety Boundaries

## Purpose
Protect sensitive data, permissions, production actions, and high-impact decisions from unsafe agent behavior.

## What To Write
Agents must not:
- Store secrets, credentials, private keys, customer records, employee records, or regulated data.
- Infer access rights from page visibility.
- Execute production actions or approve deployments.
- Rewrite authoritative policy without approval.
- Publish drafts as approved guidance.
- Remove uncertainty, disagreement, or source caveats.

Define escalation contacts for security, privacy, legal, production, and governance boundaries.

## Owner And Review Cadence
- Owner role: Security and governance owners
- Review frequency: monthly and after incidents or policy changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Treat this as a stop-sign file. If the requested action touches a boundary, ask for explicit human approval or refuse the unsafe part.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "AI Usage Policy"
description: "State how people and agents may use AI with the wiki, including review, attribution, privacy, and prohibited uses."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "AI governance and legal/privacy owners"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "quarterly and after policy changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "AI governance and legal/privacy owners"
status: "Proposal/Draft"
trust_level: "authoritative"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "quarterly and after policy changes"
audience: "all contributors and AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# AI Usage Policy

## Purpose
State how people and agents may use AI with the wiki, including review, attribution, privacy, and prohibited uses.

## What To Write
Cover:
- Allowed AI uses: drafting, summarizing, link checking, linting, template filling, and source triage.
- Prohibited uses: secrets processing, unapproved legal advice, private customer data, unreviewed policy publication, production action approval.
- Review requirements by page type.
- Disclosure or attribution requirements.
- Data-handling and redaction rules.
- Incident response if sensitive information is accidentally added.

## Owner And Review Cadence
- Owner role: AI governance and legal/privacy owners
- Review frequency: quarterly and after policy changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this to determine whether AI assistance is allowed for a task and what review is required before output becomes durable knowledge.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Governed MATM Add-On"
description: "Introduce the optional Multi-Agent Transactive Memory profile and keep it separate from the base LLM Wiki setup."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "MATM steward"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly during implementation and after memory-policy changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "MATM steward"
status: "Proposal/Draft"
trust_level: "proposal-ready"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly during implementation and after memory-policy changes"
audience: "AI system designers, memory curators, reviewers, and agent operators"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Governed MATM Add-On

## Purpose
Introduce the optional Multi-Agent Transactive Memory profile and keep it separate from the base LLM Wiki setup.

## What To Write
Use this add-on when multiple agents need to share reviewed task experience. MATM here means producer agents emit Memory Events into quarantine, a curator decides durable admission, reviewed records are indexed, and consumer agents retrieve scoped records with citations and abstention rules.

MATM is an add-on to the base LLM Wiki contract. Keep Markdown concept files under `wiki/` with required `type` frontmatter, keep `wiki/index.md` and `wiki/log.md` reserved for routing and scoped history, preserve unknown frontmatter keys, treat broken links as draft warnings and strict release failures, and partition memory by client, workspace, and project/site namespace.

MATM durable records do not replace local startup memory. Keep `.uai` / AI Memory active when present for current project truth, constraints, active blockers, project identity, handoff instructions, emergency reconnect context, and recovery posture. Consumer agents read hot local memory first, then query MATM for reviewed reusable agent experience. Do not use MATM success as a `.uai retirement path`.

This is an implementation profile, not a live service. LLMWikis publishes guidance, templates, schemas, and examples. Your deployment owns runtime authorization, storage, review, security, privacy, retention, evaluation, and incident response.

## Owner And Review Cadence
- Owner role: MATM steward
- Review frequency: monthly during implementation and after memory-policy changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this folder only when MATM is explicitly selected. Do not assume LLMWikis.org runs the memory service, grants durable write permission, or replaces local .uai startup memory.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "MATM Policy"
description: "State the governed MATM support boundary, roles, write controls, and non-claims for the local deployment."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Memory governance owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after runtime or privacy changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Memory governance owner"
status: "Proposal/Draft"
trust_level: "authoritative after local review"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after runtime or privacy changes"
audience: "memory stewards, security reviewers, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# MATM Policy

## Purpose
State the governed MATM support boundary, roles, write controls, and non-claims for the local deployment.

## What To Write
Required local policy:
- The base LLM Wiki contract still applies: concept files require `type` frontmatter, `wiki/index.md` and `wiki/log.md` stay reserved, unknown frontmatter keys are preserved, broken links are warning/strict by phase, and client/workspace/project partitioning governs scope.
- Local `.uai` / AI Memory remains the active startup continuity layer when present. MATM durable records augment it; they do not retire, disable, or make it emergency-only.
- Workers emit Memory Events only; they never write durable memory directly.
- Hidden chain-of-thought and private reasoning are not durable-memory artifacts.
- Durable records preserve compatibility with core scopes such as session, project, agent_team, and agent_repo, but deployments should map them into the six-scope practical profile when boundaries matter: global_public, global_private, project_public, project_private, agent_session, and emergency_reconnect_archive.
- Discard is a disposition sink, not a searchable durable scope.
- User identity, organization, and cross-organization memory are optional namespaced extensions with stricter privacy, access, retention, deletion, and review rules.
- Durable operations are append, update, deprecate, conflict, and discard.
- Public examples and API sketches are non-normative and do not imply a live LLMWikis service.

## Owner And Review Cadence
- Owner role: Memory governance owner
- Review frequency: monthly and after runtime or privacy changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Read before emitting, curating, indexing, or retrieving MATM records. Workers remain proposal-only, and MATM retrieval does not replace local .uai startup reads.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "MATM Wiki Decision Guide"
description: "Decide whether this LLM Wiki should adopt governed MATM and document human, agent, builder, public, protected, and operator route boundaries."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "MATM steward"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "before enabling MATM mode and after route or storage boundary changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "MATM steward"
status: "Proposal/Draft"
trust_level: "required planning gate"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "before enabling MATM mode and after route or storage boundary changes"
audience: "site owners, memory stewards, route designers, AI agents"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# MATM Wiki Decision Guide

## Purpose
Decide whether this LLM Wiki should adopt governed MATM and document human, agent, builder, public, protected, and operator route boundaries.

## What To Write
## Fit Decision

Default to an ordinary LLM Wiki. Add governed MATM only when multiple agents produce reusable operational experience that later agents need to retrieve with source status, scope, permissions, citations, abstention, review gates, and rollback evidence.

## Base LLM Wiki Contract

- Every non-reserved Markdown concept file under `wiki/` needs YAML frontmatter with a non-empty `type`.
- `wiki/index.md` and `wiki/log.md` are reserved. In a multi-project root, the root index names namespaces and project-level index/log counterparts; the root log records workspace coordination and points to project logs. Child indexes/logs carry scoped detail.
- Preserve unknown frontmatter keys and custom local metadata during migration or curation.
- Broken local links warn during draft repair and fail strict release checks.
- Partition scope by client, workspace, and project/site namespace before any MATM retrieval or durable-memory admission.

## Split-Memory Rule

- `.uai` / AI Memory is hot, local, bounded, short-term, first-load continuity when present.
- LLM Wiki is durable, reviewed, human-readable, medium/long-term knowledge.
- MATM is governed, scoped, reusable multi-agent memory.
- RAG/search is retrieval, not current operating memory.
- Archives are historical evidence, not active startup memory unless a recovery path invokes them.
- Do not retire `.uai`, write a `.uai retirement path`, stop consulting `.uai`, move all active memory out of `.uai`, date active `.uai` files as historical snapshots, or make `.uai` emergency-only because LLM Wiki/MATM ingestion, search, readback, update, or archive tests succeed.
- If `.uai/short-term-memory.uai` exists, it should describe active short-term continuity memory. If `.uai/long-term-memory.uai` exists, it should act as a pointer ledger rather than a body dump.
- If external memory becomes unavailable, recover the project basics from `.uai` first.

## Lane Map

- Human education lane: concepts, examples, and people-facing guidance.
- Agent exchange lane: public profiles, schemas, routers, compatibility checks, curation previews, and protected proposal paths.
- Builder guidance lane: decision guides, starter files, implementation levels, test fixtures, and support boundaries.
- Standards authority lane: canonical standards remain with their owning site; this file does not override them.

## Route Classes

- Public GET discovery: llms.txt, sitemap, router/profile JSON, capability registry, schema catalog, and public-safe projections.
- Public POST preview: schema validation, curation preview, retrieval preview, compatibility preview, and adoption-readiness checks. Preview is non-mutating.
- Protected submitter routes: agent registration, Memory Event submit, trajectory submit, source ingest, and wiki revision compile. Require authorization, workspace scope, idempotency, provenance, and review-pending status.
- Protected reviewer decisions: approve, request changes, reject, quarantine, revoke, and feedback. Redact private notes and payloads.
- Protected reader routes: authorized retrieval, private search, private ask, graph index, and redacted exports. Retrieval is not authorization.
- Operator evidence routes: observability, backup manifest, restore dry-run, maintenance, outbox status, and Memory Firewall reports. Expose redacted evidence only.
- Storage truth labels: distinguish memory fallback, local-test adapters, configured durable storage, worker availability, and provider status before any production claim.

## Required Surfaces

- llms.txt and sitemap entries for public guidance.
- MATM profile with version, canonical routes, status labels, capability truth labels, storage status, and explicit non-claims.
- Memory Event schema that keeps worker output proposal-only.
- Curator runbook, retrieval policy, Memory Firewall rules, adoption preview, rollback rule, and evaluation plan.
- Negative fixtures for direct durable worker write, unsupported scope, missing required fields, non-UTC timestamp, and secret-like fields.

## Anti-Patterns

- Adding MATM because the name sounds advanced.
- Letting worker agents write durable Memory Records directly.
- Treating schema validation, HMAC attestation, compatibility preview, or retrieval visibility as trust.
- Retiring `.uai`, stopping startup reads, or making `.uai` emergency-only after durable-memory readback succeeds.
- Blind-importing cognitive packets or trajectories without sandbox, provenance, review, and rollback.
- Flattening every source, schema, guide, and report into global header/footer navigation.
- Claiming live runtime, Memory Curator service, public write API, MCP/A2A support, vector database, certification, conformance, production durability, or provider-backed retrieval from handbook copy.

## Owner And Review Cadence
- Owner role: MATM steward
- Review frequency: before enabling MATM mode and after route or storage boundary changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Read before recommending MATM. Prefer an ordinary LLM Wiki unless shared reusable agent experience needs curation, retrieval, audit, and rollback; never treat MATM as .uai retirement.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Memory Scope Matrix"
description: "Define owners, contents, evidence, access, retention, promotion, demotion, retrieval, and leakage risks for MATM scopes."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Memory curator"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly during rollout"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Memory curator"
status: "Proposal/Draft"
trust_level: "reviewed guidance"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly during rollout"
audience: "curators, reviewers, retrieval engineers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Memory Scope Matrix

## Purpose
Define owners, contents, evidence, access, retention, promotion, demotion, retrieval, and leakage risks for MATM scopes.

## What To Write
Fill one row per scope:

| Scope | Owner | Purpose | Allowed contents | Prohibited contents | Minimum evidence | Access boundary | Default retention | Promotion | Demotion | Retrieval | Leakage failure |
|---|---|---|---|---|---|---|---|---|---|---|---|
| global_public | Organization steward | Reviewed shared guidance | public-safe standards and general guidance | private client or workspace data | public-safe source trace | public/shared | review cycle | broader publication review | deprecate | broad-filtered | private guidance exposed publicly |
| global_private | Organization steward | Private org memory | internal standards, protected procedures | public-only claims or secrets | private source trace | org-private | org policy | redaction and lane review | archive/deprecate | org-filtered | internal policy leaked |
| project_public | Project steward | Project-safe shared guidance | reviewed project facts safe for intended audience | unrelated project memory | project source trace | project audience | project policy | project review | deprecate | project-filtered | project rule treated as universal |
| project_private | Project steward | Private project memory | client-specific decisions, traces, runbooks | unrelated client or public data | project evidence | selected client/workspace/project | project policy | redaction and reviewer approval | archive/deprecate | selected-project-filtered | cross-project leak |
| agent_session | Run owner | Temporary task state | hypotheses, temporary observations | durable facts without curation | task trace | same run | short TTL | explicit curator promotion | discard | excluded from broad retrieval | hypothesis retrieved as fact |
| emergency_reconnect_archive | Recovery owner | Break-glass continuity | recovery packets and replay evidence | routine retrieval or public guidance | recovery manifest | permissioned read-only | recovery policy | reviewer restoration | retire | recovery-only | emergency archive used as normal memory |
| session | Run owner | Ephemeral task state | hypotheses, temporary observations | broad durable facts | task trace | same run | short TTL | explicit curator promotion | discard | excluded from broad retrieval | hypothesis retrieved as fact |
| project | Project steward | Project-bound durable memory | decisions, risks, procedures, preferences | unrelated project data | source refs | project/tenant | project policy | steward review | deprecate/archive | project-filtered | cross-project leak |
| agent_team | Team steward | Shared agent procedures | team safety rules and routing | project secrets | repeated use or review | team | monthly review | stronger review | deprecate | team-filtered | team preference treated as global |
| agent_repo | Agent owner | Specialist role memory | interface contracts, role procedures | unrelated team policy | owner review | role/agent | owner review | owner review | deprecate | role-filtered | specialist rule used by wrong agent |
| discard | Curator | Disposition sink | unsafe or unsupported events | searchable memory | reason where allowed | audit only | policy | none | none | not retrievable | discarded event indexed |

## Owner And Review Cadence
- Owner role: Memory curator
- Review frequency: monthly during rollout
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use this to route Memory Events before durable admission. Do not add a fifth core scope silently.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Memory Taxonomy"
description: "Define the nine core MATM memory kinds and their evidence, retrieval, promotion, and review rules."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Memory taxonomy owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly during rollout"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Memory taxonomy owner"
status: "Proposal/Draft"
trust_level: "reviewed guidance"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly during rollout"
audience: "curators, agents, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Memory Taxonomy

## Purpose
Define the nine core MATM memory kinds and their evidence, retrieval, promotion, and review rules.

## What To Write
Core kinds: fact, decision, preference, risk, procedure, hypothesis, evidence, deprecation, conflict.

Rules:
- A hypothesis is not a fact.
- Evidence supports a claim but is not automatically the claim.
- A decision includes context, authority, rationale, date, and effective scope.
- A preference is scoped and should not become universal policy.
- A procedure needs trigger, preconditions, steps, validation, stop conditions, and rollback or escalation.
- A deprecation references the superseded record.
- A conflict preserves credible sides until resolution.

For each kind, add definition, positive example, counterexample, evidence requirement, permitted scopes, prohibited promotions, retrieval treatment, review trigger, and retention note.

## Owner And Review Cadence
- Owner role: Memory taxonomy owner
- Review frequency: monthly during rollout
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Classify events using exactly the nine core kinds unless an extension is explicitly reviewed.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Memory Emission Contract"
description: "Specify how worker agents emit proposal-only Memory Events."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Agent platform owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "after agent runtime changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Agent platform owner"
status: "Proposal/Draft"
trust_level: "authoritative after local review"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "after agent runtime changes"
audience: "producer agents, orchestrators, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Memory Emission Contract

## Purpose
Specify how worker agents emit proposal-only Memory Events.

## What To Write
Worker emission contract:
- Emit `schema_version`, `event_id`, UTC `emitted_at`, producer identity, task id, artifact summary, memory kind, suggested scope, disposition hint, sensitivity, confidence, evidence refs, provenance, and review hints.
- Set `worker_write_mode` to `proposal_only`.
- Set `durable_write_allowed` to `false`.
- Include observable artifact references such as trajectories, tool results, plans, outcome summaries, tests, or failure analyses.
- Exclude hidden chain-of-thought, private reasoning, secrets, credentials, and raw regulated data.
- Route every event to quarantine before curation.

## Owner And Review Cadence
- Owner role: Agent platform owner
- Review frequency: after agent runtime changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Workers must use this contract before emitting Memory Events and must never bypass the curator.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Memory Curator Runbook"
description: "Define the admission pipeline that turns Memory Events into reviewed Memory Records or discard decisions."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Memory curator"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after incidents"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Memory curator"
status: "Proposal/Draft"
trust_level: "authoritative after local review"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after incidents"
audience: "curators, reviewers, runtime engineers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Memory Curator Runbook

## Purpose
Define the admission pipeline that turns Memory Events into reviewed Memory Records or discard decisions.

## What To Write
Nine-step pipeline:
1. Schema admission.
2. Safety and redaction.
3. Source and provenance check.
4. Scope routing.
5. Kind classification.
6. Evidence sufficiency.
7. Deduplication and idempotency.
8. Conflict analysis.
9. Disposition and audit.

Durable operation decisions are append, update, deprecate, conflict, or discard. Discard records the rejection decision and reason; it is not a durable memory scope.

High-risk public, sensitive, cross-scope, legal, security, privacy, production, or conflict cases require human or specialist review before durable admission.

## Owner And Review Cadence
- Owner role: Memory curator
- Review frequency: monthly and after incidents
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use before durable admission. If curator ownership is missing, keep events in quarantine.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "Curation Decision Table"
description: "Provide a compact table for append, update, deprecate, conflict, and discard decisions."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Memory curator"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "after policy changes"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Memory curator"
status: "Proposal/Draft"
trust_level: "reviewed guidance"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "after policy changes"
audience: "curators and auditors"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# Curation Decision Table

## Purpose
Provide a compact table for append, update, deprecate, conflict, and discard decisions.

## What To Write
| Operation | Use when | Required audit | Not this |
|---|---|---|---|
| append | New reviewed memory is useful and in scope. | event id, evidence, scope, kind, owner | worker direct write |
| update | Existing record should change. | prior record, source refs, reviewer | silent overwrite |
| deprecate | Record is superseded, stale, unsafe, or retired. | replacement or reason | routine delete |
| conflict | Credible incompatible claims must remain visible. | both sides and reviewer | fake consensus |
| discard | Event is malformed, unsafe, unauthorized, redundant, unsupported, or out of scope. | reason where policy permits | searchable durable memory |

## Owner And Review Cadence
- Owner role: Memory curator
- Review frequency: after policy changes
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use exact operation names. Propose or queued are workflow states, not durable write primitives.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "MATM Retrieval Policy"
description: "Define filtering, ranking, citation, personalization, caching, and abstention rules for consumer agents."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Retrieval owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after retrieval incidents"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Retrieval owner"
status: "Proposal/Draft"
trust_level: "reviewed guidance"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after retrieval incidents"
audience: "retrieval engineers, consumer agents, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# MATM Retrieval Policy

## Purpose
Define filtering, ranking, citation, personalization, caching, and abstention rules for consumer agents.

## What To Write
Retrieval must filter by authorization, scope, sensitivity, status, evidence, conflict state, and deprecation state before similarity ranking.

Consumer agents read local `.uai` / AI Memory / Project Handoff first for current truth, constraints, active blockers, and recovery posture. MATM retrieval supplies reviewed reusable agent experience after that startup context is loaded.

Rank with relevance, task state, producer reliability, evidence quality, freshness, conflict status, and consumer fit. Do not rank a hypothesis as fact. Return citations and warnings. Abstain on missing source/status/scope/authorization metadata, cross-scope leakage risk, stale-only matches, deprecated-only matches, or unresolved high-impact conflicts.

## Owner And Review Cadence
- Owner role: Retrieval owner
- Review frequency: monthly and after retrieval incidents
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Retrieve reviewed scoped records with citations after local .uai / AI Memory startup context is read. Abstain when metadata or authority is missing.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "MATM Security And Privacy"
description: "Define controls for memory poisoning, cross-scope leakage, redaction, access control, retention, and public export."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Security and privacy owners"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "monthly and after incidents"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Security and privacy owners"
status: "Proposal/Draft"
trust_level: "authoritative after local review"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "monthly and after incidents"
audience: "security, privacy, curators, runtime owners"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# MATM Security And Privacy

## Purpose
Define controls for memory poisoning, cross-scope leakage, redaction, access control, retention, and public export.

## What To Write
Required controls: authenticated producer identity, capability-scoped authorization, per-scope access control, tenant/project isolation, schema validation, sanitization, prompt-injection handling, evidence allowlists, rate limits, idempotency, immutable audit records, human-review gates, retention, deprecation propagation, incident response, privacy deletion workflow, and public-export review.

Policy-based redaction is not a formal privacy guarantee. Differential privacy, cryptographic provenance, federated consensus, and inter-organization memory are optional advanced work unless implemented and validated locally.

## Owner And Review Cadence
- Owner role: Security and privacy owners
- Review frequency: monthly and after incidents
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Stop on secrets, unclear consent, cross-scope leakage, poisoning risk, or unsupported public export.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "MATM Evaluation Plan"
description: "Separate research evidence, curator simulation, and deployment-specific field evidence."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Evaluation owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "after experiments and before rollout milestones"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Evaluation owner"
status: "Proposal/Draft"
trust_level: "reviewed guidance"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "after experiments and before rollout milestones"
audience: "evaluators, curators, deployment owners"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# MATM Evaluation Plan

## Purpose
Separate research evidence, curator simulation, and deployment-specific field evidence.

## What To Write
Track three evidence classes:
1. MATM empirical retrieval evidence.
2. Memory Curator theoretical or simulation evidence.
3. Deployment-specific field evidence.

Measure task success, interaction steps, retrieval hit rate, precision/recall, routing accuracy, scope leakage, evidence sufficiency, duplicates, conflict precision/recall, stale retrieval, deprecated-record retrieval, privacy leakage, poisoning resistance, curator false admission/rejection, latency, token usage, storage growth, review burden, cost, availability, and cache invalidation correctness.

## Owner And Review Cadence
- Owner role: Evaluation owner
- Review frequency: after experiments and before rollout milestones
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Do not combine paper results, simulation, and local deployment metrics into one production-readiness claim.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.

---
type: "Reference"
title: "MATM Migration Checklist"
description: "Migrate direct-write agent memory toward proposal-only, curator-governed memory."
resource: ""
tags: []
timestamp: "2026-07-02T00:00:00Z"
llmwiki_status: "draft"
llmwiki_owner: "Migration owner"
llmwiki_source_status: "starter-template"
llmwiki_agent_use: "read-cite-propose"
llmwiki_review_cycle: "during migration milestones"
llmwiki_created_at: "2026-07-02T00:00:00Z"
llmwiki_created_by: "LlmWikis.org starter bundle"
llmwiki_modified_at: "2026-07-02T00:00:00Z"
llmwiki_modified_by: "LlmWikis.org starter bundle"
llmwiki_version: "3.2.0"
owner: "Migration owner"
status: "Proposal/Draft"
trust_level: "reviewed guidance"
source_status: "starter-template"
canonical_url: "/"
routing_summary: "Explain why this page should be loaded and what task it routes."
source_trace: []
checksum_or_source_identity: ""
discovery_surface: "llm-wiki/INDEX.md"
last_reviewed: "YYYY-MM-DD"
review_cycle: "during migration milestones"
audience: "runtime owners, curators, reviewers"
sensitivity: "internal"
agent_use: "allowed-with-citation"
namespace: "wiki/global/"
workspace_scope: "global"
human_review_required_for_updates: true
typed_relations:
  part_of:
    - "llm-wiki/INDEX.md"
---

# MATM Migration Checklist

## Purpose
Migrate direct-write agent memory toward proposal-only, curator-governed memory.

## What To Write
Migration sequence:
1. Freeze or gate direct durable writes.
2. Inventory stores and writers.
3. Classify sensitive and private content.
4. Separate session and durable memory.
5. Confirm local `.uai` / AI Memory startup memory remains active and is not retired, disabled, or made emergency-only by the durable-memory rollout.
6. Backfill provenance where possible.
7. Add Memory Event emission.
8. Add quarantine/proposal storage.
9. Run curator decisions in shadow mode.
10. Measure routing, leakage, evidence, conflicts, latency, and downstream utility.
11. Enable approved durable writes by curator only.
12. Re-index only reviewed records.
13. Add revalidation, deprecation, retention, and audit cadence.

## Owner And Review Cadence
- Owner role: Migration owner
- Review frequency: during migration milestones
- Human review required for updates: true
- Mark this file `needs-review` if the owner is unknown, the review date is stale, or the content conflicts with a more authoritative page.

## How An LLM Should Use This File
Use for audits of existing agent-memory systems before enabling governed MATM writes. Do not use migration to retire, replace, date, or move all active memory out of local .uai startup memory.

## Update Prompt
When updating this file, preserve sources, note uncertainty, update the review date only after human review, and add related pages that a reader or agent should inspect next.